Destructive Creation of New Invasive Technologies: Generative Artificial Intelligence Behaviour

Abstract

This study proposes a new concept that explains a source of technological change: The invasive behaviour of general purpose technologies that breaks into scientific and technological ecosystems with accelerated diffusion of new products and processes that destroy the usage value of all units previously used. This study highlights the dynamics of the invasive destruction of new path-breaking technologies in driving innovative activity. Invasive technologies conquer the scientific, technological, and business spaces of alternative technologies by introducing manifold radical innovations that support technological, economic, and social change. The proposed theoretical framework is verified empirically in new technologies of neural network architectures, comparing transformer technology (a deep learning architecture having unsupervised and semi-supervised algorithms that create new contents and mimic human ability, supporting Generative Artificial Intelligence) to Long Short-Term Memory (LSTM) and Recurrent Neural Networks (RNNs). Statistical evidence here, based on patent analyses, reveals that the exponential growth rate of transformer technology over a period of five years (2020–2024) is 45.91% more than double compared to the alternative technologies of LSTM (21.17%) and RNN (18.15%). Moreover, the proposed invasive rate in technological space shows that is very high for transformer technology at the level of 2.2%, whereas for LSTM it is 1.39% and for RNN it is 1.22% over 2020–2024, respectively. Invasive behaviour of drastic technologies is a new approach that can explain one of the major causes of global technological change and this scientific examination here significantly contributes to our understanding of the current dynamics in technological evolution of the Artificial Intelligence technology having high industrial impacts on the progress of human society.

1. Introduction and Scientific Goal

The goal of this study is to suggest a new concept that drives technological, economic, and social change: The invasive behaviour of new technologies that is a hardly known characteristic. Invasion is anything that breaks into a space, occupying it or spreading in large quantities in the short run. In nature, there are different aspects of invasion: in botany, invasive plants invade lands and human habitats [1,2]; in biology, invasive organisms are not indigenous to a particular area and cause environmental harm [3]; and in medicine, invasive cancer navigates in different tissue microenvironments of the body [4,5].

Invading organisms in ecology explain the main interactions in the total environment [6]. However, the invasive behaviour in the study of technologies and innovations is a characteristic unknown but its examination is necessary for clarifying new properties of technological diffusion, evolution, and change. For the first time, this study analyses these aspects with a broad analogy with fundamentals in biology and a theoretical framework of generalized Darwinism [7,8]. The proposed scientific concept of invasiveness in the economics of innovation is studied to explain the behaviour of new technologies and clarify the current dynamics of technological change. The suggested invasive behaviour of new technologies is intended to destroy established technologies, to occupy their space, and become the dominant technology to drive innovative activity supporting technological, economic, and social change. In particular, invasive destruction can be a main characteristic of new general purpose technologies that generate a technological revolution (or technological paradigm shift) and related innovation avenues that pave different interactions with manifold technologies by “expanding the adjacent possible” in science and technological fields [8,9,10,11,12,13,14,15,16,17].

What this study adds to innovation studies is the concept of technology invasiveness that lays the foundation for a better theory that explains the evolutionary behaviour of new technologies driving technological and social change. Hence, this study offers, with the concept of invasiveness, a new exploration of technological behaviour directed to clarify technological evolution for progress in human society.

In the presence of this main scientific goal, the research questions are as follows:

• What are the basic characteristics of invasive technologies compared to alternative technologies?
• How long does it take for invasive technologies to expand in the technological space and become dominant?
• How can the invasive behaviour of new technologies be measured?

This study confronts these research questions by developing the technological invasiveness theory, which endeavours to explain new sources of technological change in regimes of rapid change. Statistical evidence, which focuses on radical technologies generated by knowledge creation [18] in Generative Artificial Intelligence (AI), endeavours to explain the invasive behaviour showing aspects of the expansion in technological space that destroys existing products and changes the dynamic capabilities [19,20]. This proposed concept of invasive technologies is especially relevant in innovation-based ecosystems that generate the multifunctional loop of technological drivers given by new knowledge creation–destruction of existing competencies and –further “creative destruction” (Figure 1). The proposed approach can support both better theory to explain technological change for human progress and best practices of innovation management.

2. Theoretical Background

One of the fundamental problems in technological studies is to explain the behaviour of drastic technology directed to economic and social change [21,22,23,24,25,26,27,28]. The main theoretical framework is by Christensen [29,30,31] on disruptive innovations that cause a relevant change in firms, industries, and markets. Disruptive technology creates new sectors and business models that change the market structure and how products and services are yielded and consumed. The characteristics of destructive technology, which generates radical innovations of new products and/or processes, are high technical and/or economic performance directed to destroy the usage value of established technologies and related products/processes to expand its market share [29,30,31,32]. Calvano [33] maintains that “Destructive Creation” is the deliberate introduction of new and improved generations of products that destroy, directly or indirectly, current products, inducing consumers to change their habits with consequential economic and social change [34,35,36]. Adner ([37], pp. 668–669) claims that “Disruptive technologies … introduce a different performance package from mainstream technologies” [35,38]. Abernathy and Clark ([39], pp. 4ff and pp. 12–13) clearly mention that “Innovation that disrupts and renders established technical and production competence obsolete … is … labelled ‘Revolutionary’. It thus seems clear that the power of an innovation to unleash Schumpeter’s ‘creative destruction’ must be gauged by the extent to which it alters the parameters of competition, as well as by the shifts it causes in required technical competence”. Christensen [30] argues that disruptive technology has specific properties: (a) higher technological performance and (b) generation of new products/processes that satisfy the needs that are demanded by mainstream markets. Christensen et al. [29] also claim that disruptive technologies can be generated by small firms with fewer resources that successfully open new industries and/or challenge established incumbent businesses in current sectors (e.g., OpenAI Inc., established in December 2015 in the city of San Francisco-California, U.S.A., which released in November 2022 the first ChatGPT version that used the GPT-3.5 model). Innovative firms, generating disruptive technologies and innovations, grow more rapidly than other ones ([39,40], Tushman and Anderson [41], p. 439). Christensen’s [30] approach also shows that disruptive technologies or innovations (these terms can be used here interchangeably) generate significant shifts in economic and social systems [42]. In general, the main technological shifts embody path-breaking technologies that are competence-destroying [41,43]. Moreover, disruptive innovations undermine the competencies and complementary assets of existing producers, and change the behaviour of firms and consumers, fostering economic and social changes [31,44,45]. The development and diffusion of disruptive innovation create and sustain competitive advantage in firms and nations [46,47]. Disruptive technology can also support a new technological paradigm that generates different technological trajectories driving processes of substitution in new techniques for the established ones and, as a consequence, affect the evolution of manifold inter-related technologies [48,49].

What this study adds to the existing literature is the concept of invasive technology as a larger and more powerful technological system than disruptive technologies that can explain the rapid technological change in modern economies of hyper-connected ecosystems.

The next section presents the research philosophy, methodology, and study design to structure theory and empirical evidence in invasive technologies.

3. Methods of Research

3.1. Rationale and Research Philosophy of This Study

The proposed theoretical framework here is developed with a perspective of generalized or universal Darwinism [50,51,52]. Hodgson ([53], p. 260) maintains that “Darwinism involves a general theory of all open, complex systems”. Hodgson and Knudsen [7] suggest that a generalization of the Darwinian concepts can explain how a complex system evolves [53,54,55]. Darwinian principles (“Generalized Darwinism”) can assist in explaining the multidisciplinary nature of scientific and technological processes [7,8,51,52,56]. In fact, the heuristic principles of “Generalized Darwinism” can clarify aspects of scientific and technological development considering analogies between evolution in biological systems and in scientific–technological systems [57,58]. Arthur [59] argues that Darwinism approach can support understanding of the dynamics of science and technology using similar concepts and methods to understanding the development of species in natural ecosystems. Kauffman and Macready ([11], p. 26) state that “Technological evolution, like biological evolution, can be considered a search across a space of possibilities on complex, multipeaked ‘fitness,’ ‘efficiency,’ or ‘cost’ landscapes”. Schuster ([60], p. 8) shows the similarity between technological and biological evolution, such as finite lifetimes in technologies like biological organisms, differentiation of elements, etc. In general, technological and scientific evolution, like biological evolution, displays novelty, radiations, stasis, survival, adaptation, and extinctions [11,61,62]. However, unlike ecology, the invasive behaviour in technological studies is unknown but it can be basic to explain important characteristics of technological change. The theoretical framework of “Generalized Darwinism” [7], just described, can frame a broad analogy between biological evolution and similar aspects in science and technology processes that provides a logical structure of scientific inquiry to analyse and explain proposed invasive behaviour of new technologies that generate technological change [9,10,63,64,65].

3.2. Basic Concepts of Proposed Theoretical Framework in Invasive Technologies

• Invasion is a behaviour that bursts and spreads in space, occupying the position of other elements in the system in the short or long run.
• Invasive technologies can replace other technologies in the short run, producing a lot of innovations that have the potential to spread in different scientific and industrial sectors leading to significant technological, economic, and social change in ecosystems.

Postulations

• Invasive technologies have adaptive behaviour in different ecosystems and eliminate less suitable technologies.
• Invasive technologies are a driver of technological and social change.
• Invasive behaviour ⇒ macro-technological evolution.

Predictions of the theory of invasive technologies

• The short-run rate of growth in invasive technology is at least twofold the rate of growth in alternative technologies.
• Invasive technology (i) is better adapted than alternative technologies (j) in ecosystem S, if and only if (i) is able to produce and spread new innovations in S than (j) over time and space.

3.3. Research Setting to Analyse the Invasive Behaviour of New Technologies in Neural Network Architectures

The predictions of the proposed theory of invasive technologies are verified empirically in some main technologies in Artificial Intelligence (AI) with a comparative analysis. Figure 2 shows chronological evolution of main neural network architecture supporting Generative Artificial Intelligence, indicating the acronym and year of origin.

• The first technology under study to explain invasive behaviour is Recurrent Neural Networks (RNNs), which were introduced in 1985 [66]. RNNs were designed to handle sequential data, making them suitable for tasks like language modelling and speech recognition. RNNs are powerful technologies but they have limitations, such as slow training, poor retention of old connections, and struggling with long-term dependencies.
• The second technology under study is Long Short-Term Memory (LSTM), which was introduced in 1997 [67]. LSTMs are a type of Recurrent Neural Network (RNN) designed to address the issue of long-term dependencies by using memory cells to retain information over longer sequences. Although LSTMs are better than traditional RNNs at handling long-term dependencies, they still struggle with very long sequences because of their sequential nature.
• The third radical technology under study here is transformer (TFR) architecture, which is a new type of neural network, described by Vaswani et al. [68]. Unlike Recurrent Neural Networks (RNNs), the new architecture of transformer technology is based on three powerful elements: (a) self-attention; (b) positional embeddings; and (c) multi-head attention [69,70]. Moreover, unlike LSTMs, which are difficult to parallelize, transformers are highly parallelizable because they process all tokens in the input sequence simultaneously, making them much faster to train on large datasets. Transformer architecture from 2018 is developing pretrained language models (Generative Pretraining Transformers, GPTs series), such as GPT-1 in 2018 (using a transformer architecture to predict the next word in a sentence, with a parameter count of about 117 million; [71,72]), GPT-2 in 2019, GPT-3 in 2020, GPT-4 in 2023, which is capable of generating human-like content with better performance, and GPT-4o in 2024, which has a parameter count of more than 2 trillion [71,72,73,74]. Other main innovations generated by transformer technology are Google’s Bidirectional Encoder Representations from Transformers (BERTs) model [75], Microsoft copilot [76], etc. A wide range of applications in this new technology exist [77,78], including machine translation, document summarization, document generation, named entity recognition, biological sequence analysis, writing computer code based on requirements expressed in natural language, video understanding, computer vision, protein folding applications, etc.

3.4. Study Design

A comparative analysis of the behaviour of transformer (TFR) technologies (a new type of deep learning architecture used in Generative Artificial Intelligence models) is carried out with two technologies previously described: Long Short-Term Memory and Recurrent Neural Networks. This study focuses on two periods in which these technologies emerged and spread:

• 2014–2024 (last year with full data available), this period indicates a medium run
• 2020–2024 (last year with full data available), this period indicates a short run.

3.5. Measures and Sources of Data

The predictions of the proposed theory in invasive technologies are verified with patent analysis (patents in the economics of innovation are a main proxy of inventions and innovations [79]). Data are from the online library database Scopus [80], downloaded on 12 March 2025. The year 2024 is the last year with full data available.

3.6. Logic Structure of the Search String to Gather Data

Technological space includes all technologies Ti (i = 1, …n) having characteristics to solve specific problems and/or satisfy needs in human societies, such as deep learning models that involve the training of artificial neural networks to recognize patterns and make decisions. The search strategy here is to detect patents in the new technology Ti understudy in the technological space of deep learning (set A) and then to consider the complementary set A^C = B given by set A without the new technology Ti under study. This strategy shows the dynamics of new technology compared to all alternative technologies.

Using the set theory, set A of deep learning models = {T₁, T₂, …, T_n}, a new technology (e.g., a specific artificial neural network) is defined as T_i ⊂ A

T^C = B, i.e., B is A − { T_i }, then, T_i ∪ {B= T^C }= set A= {T₁, T₂, …, T_n}.

The library database Scopus [80] was used to search for patents updated over time. The search strings, directed to detect patents of technologies under study for comparative analysis, are based on a combination of specific keywords and Boolean operators (AND, AND NOT), which were inserted into the search box of the search engine Scopus [80] as follows:

For the new technology of transformers, the search queries were as follows (on 12 March 2025; [80]):

T1) query: ((TITLE-ABS-KEY(“deep learning”) AND TITLE-ABS-KEY(transformer))).

B1) query: ((TITLE-ABS-KEY(“deep learning”) AND NOT TITLE-ABS-KEY(transformer)))

Of course, T1∪B1=A1

The behaviour of this new technology was compared to two alternative technologies:

RNN:

T2) query: ((TITLE-ABS-KEY(“deep learning”) AND TITLE-ABS-KEY(“Recurrent Neural Network”)))

B2) query: ((TITLE-ABS-KEY(“deep learning”) AND NOT TITLE-ABS-KEY(“Recurrent Neural Network”)))

Of course, T2∪B2=A2

LSTM:

T3) query: ((TITLE-ABS-KEY(“deep learning”) AND TITLE-ABS-KEY(“Long Short-Term Memory”)))

B3) query: ((TITLE-ABS-KEY(“deep learning”) AND NOT TITLE-ABS-KEY(“Long Short-Term Memory”)))

Of course, T3∪B3=A3

Hence, T is the set that includes cases of new technology with invasive behaviour. The set B =T^C (1, 2, or 3 according to the three technologies) includes the domain of technologies that have been predated by new invasive technology.

As a consequence, T∪B=A

3.7. Samples

This study considers the following sample of patents summarized in

Table 1. Data of patents in new technologies (set T) and the set B.

	Transformer (T1)	Complement Set B1	RNN (T2)	Complement Set B2	LSTM (T3)	Complement Set B3
Patents 2014–2024	13,984	191,879	40,790	165,073	22,821	183,042

Note: T∪B=A of deep learning models = {T₁, T₂, …, T_n}.

, which was detected using the previous logic of search strings in the dataset of Scopus [80].

3.8. Modelling and Data Analysis Procedures

One significant way to understand the invasive behaviour of new transformer technologies (TFR) is to estimate the rates of diffusion in technological space compared to alternative technologies, i.e., RNN and LSTM.

i.

Rates of growth

The temporal development of these technologies was analysed with an exponential rate of growth: r. In this case, the function of patent development is exponential:

$${Patents}_t{=Patents}_0{e}^{rT}$$

(1)

Hence, ${\displaystyle \frac{P_{atentst}}{P_{atents0}}}=e^{rT}$, where the term e is the base of the natural logarithm (2.71828…) as follows:

$$Log{\displaystyle \frac{{Patents}_t}{{Patents}_0}}=rT$$

(2)

$$r={\displaystyle \frac{Log\frac{{Patents}_t}{{Patents}_0}}{T}}$$

(3)

where

• r = the rate of exponential growth of technology from 0 to t
• P₀ is the patents to time 0
• P_t is the patents to time t
• T = t − 0

This study considers the rate of exponential growth in the medium-period of T = 10 years (2024–2014) and the short-period of T = 5 years (2024–2020).

The data, transformed into a log scale, are also represented with area charts for a comparative analysis of the invasive behaviour of new technologies under study in the technological space of alternative technologies.

Moreover, for robustness, the rate of growth considering the period T = 5 years and the annual rate of growth were calculated as follows:

$${\phi }_{ T= 5 years}={\displaystyle \frac{Pt-P(t-5)}{T=5}}$$

(4)

$${\phi }_{ annual}={\displaystyle \frac{{\phi }_{ T}}{T}}$$

(5)

The temporal difference (Δ) in the rates of growth in these technologies also provides the main information:

$$∆=rt-r(t-5)$$

(6)

ii.

Temporal aspects of invasive technologies

Trends of invasive technology i (transformers, RNN or LSTM) at t are analysed with the following log-linear model:

log y_i,t = a + b time + u_i,t

(7)

y_i,t = patents of invasive technologies i at t

t = time; period under study is 10 years or 5 years

a = constant; b = coefficient of regression; u_i,t = error term.

Mutatis mutandis, the log-linear model is also performed with the complement set of deep learning models without the technologies under study: transformers, RNN, or LSTM.

The dependent variable is logNOTy_i,t

iii.

Spatial aspects of invasive technologies based on technological substitution

These new technologies are also analysed with a model based on spatial aspects of technological evolution in which the number of patents in new technology (Y) is a function of the number of patents in previous technology (X), considering an approach of technological substitution. This model provides the relative rate of technological evolution of new technology compared to previous technology [24]. In fact, the adoption of new technology is associated with the nature of some comparable previous technology in use, which is destroyed. In fact, the evolution and diffusion of new technology with destructive effects do not take place in isolation, but it is a process of substitution of new technology for the old one. Pistorius and Utterback [81] argue that emerging technologies often substitute for more mature technologies. Porter [47] considers substitutes as one of the forces in his model of industrial competition for the competitive advantage of firms and nations. Fisher and Pry ([49], p. 75) argue that technological evolution consists of substituting new technology for the old one, such as the substitution of coal for wood, hydrocarbons for coal, etc. Fisher and Pry [49] modelled the evolution of a new product or process by becoming a substitute for a prior one (cf., Utterback et al., [27], p. 2). Fisher and Pry ([49], p. 88) state that: “The speed with which a substitution takes place is not a simple measure of the pace of technical advance … it is rather a measure of the imbalance in these factors between the competitive elements of the substitution”.

In this context, the structure of the model is as follows:

Let Y(t) be the extent of advances of new technology (e.g., transformers) at the time t, measured with patents, and X(t) be the patents underlying the advances of previous technology (e.g., RNN). Suppose that both X and Y evolve according to an S-shaped pattern that can be represented with a differential equation of the logistic function. The logistic model has a symmetrical S-shaped curve with a point of inflection at 0.5 K. After some mathematical transformations [24], the differential equation of logistic function is a simple linear relationship (log–log model):

$$\mathit{log}Y=\mathit{log}A+B \mathit{log}X$$

(8)

A = constant; $B$ = evolutionary coefficient of growth that measures the technological substitution of new technology Y (patents) in relation to patents in previous technology X.

The coefficient of relative growth B in model (8) indicates different pathways of technological evolution in new technology:

• B < 1 indicates that the new technology Y evolves at a lower relative rate of change than X (previous technology), thus slowing down substitution over the course of time.
• $B=1$ indicates that Y and X have a proportional evolution over time (proportional substitution).
• B > 1 indicates that Y evolves at a greater relative rate of change than X; the technological system in the new technology Y has an accelerated evolution of patents compared to patents in X over the course of time, generating a process of allometric growth that supports the accelerated technological substitution.

iv.

Model of technological invasion of new technologies and metrics of technological invasiveness

In order to measure the rate of technological invasiveness, the following model is designed from a perspective of the interaction between predator (new technologies, e.g., transformer) and prey (other alternative technologies in the domain of deep learning), where the growth rate of new technologies inhibits the growth rate of other alternative technologies (Pistorius and Utterback [81], p. 74). Farrell [82,83] used a model based on Lotka–Volterra equations to examine a predator–prey relationship between emerging technology and established technology (e.g., nylon versus rayon tire cords, telephone versus telegraph usage, etc.).

The following model provides a main measure of technological invasiveness with the coefficient I. First, Y(t) is the extent of advances of new technology (predator–invader) at the time t, measured with patents, and D(t) is the patents underlying the advances of other technologies (prey) in the technological space of deep learning models. The model here also assumes that both D (prey in invaded technological space) and Y (predator–invader technology) evolve according to an S-shaped pattern that can be represented with a differential equation of the logistic function. Mathematical transformations of logistic function [24] also generate a linear relationship (log–log model):

$$\mathit{log}Y\text{'}=\mathit{log}A\text{'}+I \mathit{log}D$$

(9)

The model focuses on a period of 10 and 5 years to assess the effective rate of invasion in the medium- and short- periods.

A’ = constant

I = evolutionary coefficient of invasiveness that measures how (velocity) the new technology Y invades the technological space of alternative technologies D over time.

The coefficient of relative invasiveness I in the model (9) indicates different pathways of technological invasion in new technology as follows:

• I < 1 indicates that the new predator technology Y invades the technological space at a slow rate of diffusion compared to the rate of alternative technologies D over the course of time.
• I = 1 indicates that the new predator technology Y invades the technological space at a proportional diffusion rate to the rate of growth in alternative technologies D over the course of time.
• I > 1 indicates that the new predator–invader technology Y has an accelerated rate of invasion in technological space compared to the rate of growth in alternative technologies (D = prey); hence, the new predator–invader technology Y has an accelerated rate of invasion in the technological space.

The estimation of models with the Ordinary Least Squares (OLS) method was used to determine the unknown parameters. Statistical analyses were performed with IBM SPSS Statistics 26 ^®.

4. Statistical Analyses to Verify Theory of Invasive Technologies

Area graphs in Figure 3, Figure 4 and Figure 5 show, over time, the invasive dynamics of new technologies in the technological space of alternative technologies over the 2014–2024 period.

Table 2. Different rates of growth in technologies and complementary sets of alternative technologies based on patents over time.

NEW TECHNOLOGIES	Exponential Rate (2014–2024)	Exponential Rate (2014–2018)	Exponential Rate (2016–2020)	Exponential Rate (2020–2024)	G = (P2024-P2020) /P2020	Annual G
Transformer	79.69	87.13	77.63	45.91	8.93	1.79
Domain of alternative technologies to transformers	67.62	96.05	71.35	12.65	0.88	0.18
RNN	81.36	116.1	93.28	18.15	1.48	0.30
Domain of alternative technologies to RNN	66.69	93.36	68.72	13.89	1.00	0.20
LSTM	87.25	125.53	100.54	21.17	1.88	0.38
Domain of alternative technologies to LSTM	67.32	94.30	69.95	13.94	1.01	0.20

shows the invasive behaviour of new technologies, in particular in the short run of five years (2020–2024) when transformer technology emerged and has an exponential growth rate of 45.9%, which is higher than LSTM (21.2%) and RNN (18.2%). This invasive behaviour was also present when LSTM emerged compared to other technologies. Period and annual rates of growth, G, are in the last two columns of Table 2. The total share of patents in the technologies and technological domains, including alternative technologies in deep learning and rate differences from 2020 to 2024, confirm the results just mentioned (

Table 3. The total share of patents in the technologies and complementary technological domains in deep learning over time (sum=1 per year and technology).

Years	Transformer	Technological Space of Alternative Technologies to Transformers	RNN	Technological Space of Alternative Technologies to RNN	LSTM	Technological Space of Alternative Technologies to LSTM
2014	0.040	0.960	0.060	0.940	0.020	0.980
2015	0.009	0.991	0.027	0.973	0.000	1.000
2016	0.019	0.981	0.062	0.938	0.022	0.978
2017	0.023	0.977	0.099	0.901	0.041	0.959
2018	0.026	0.974	0.166	0.834	0.089	0.911
2019	0.028	0.972	0.172	0.828	0.083	0.917
2020	0.026	0.974	0.183	0.817	0.094	0.906
2021	0.036	0.964	0.192	0.808	0.102	0.898
2022	0.056	0.944	0.194	0.806	0.112	0.888
2023	0.084	0.916	0.214	0.786	0.120	0.880
2024	0.122	0.878	0.217	0.783	0.130	0.870
Rate Difference 2024–2020	9.652		3.399		3.578

).

4.1. Pattens of Temporal and Morphological Change in Invasive Technologies

Table 4. Estimated relationships of patents as a function of time.

	Dependent Variable: Patents Concerning Different Technologies
T = 10 Years	Coeff. b’1	Stand. Beta	Constant a’	F	R2	Period
Log Transformers10 years	0.85 ***	0.99	−1703.89 ***	211.41 ***	0.96	(2014–2024)
Log NOT Transformers10 years	0.67 ***	0.94	−1340.88 ***	69.11 ***	0.89	(2014–2024)
Log RNN10 years	0.85 ***	0.94	−1718.78 ***	68.66 ***	0.88	(2014–2024)
Log NOT RNN10 years	0.65 ***	0.94	−1312.48 ***	72.52 ***	0.89	(2014–2024)
Log LSTM 10 years	0.86 ***	0.94	−1720.38 ***	56.44 ***	0.88	(2014–2024)
Log NOT LSTM10 years	0.64 ***	0.93	−1289.37 ***	53.02 ***	0.87	(2014–2024)
T = 5 years	Coeff. b’1	Stand. Beta	Constant a’	F	R2	Period
Log Transformers5 years	0.57 ***	0.99	−1145.49 ***	754.12 ***	0.99	(2020–2024)
Log NOTTransformers5 years	0.23 ***	0.93	−445.66 ***	25.48 **	0.86	(2020–2024)
Log RNN5 years	0.29 **	0.96	−583.68 **	49.42 **	0.93	(2020–2024)
Log NOT RNN5 years	0.24 **	0.94	−464.21 **	32.85 **	0.89	(2020–2024)
Log LSTM 5 years	0.33 **	0.97	−667.22 ***	54.74 **	0.93	(2020–2024)
Log NOT LSTM5 years	0.24 **	0.95	−466.13 **	33.50 **	0.89	(2020–2024)

Note: The explanatory variable is time in years. The period in the last column: The first year indicates the first year under study; the second year is 2024, which is the last available year. *** significant at 1‰. ** significant at 1%. F is the ratio of the variance explained by the model to the unexplained variance. R² is the coefficient of determination.

shows the estimated relationships of patents as a function of time. Regression analyses over a period of five years (2020–2024) show that log-linear models estimated provide robust statistical results. The F value is significant (p-value < 0.001 or <0.05), and the coefficient R² is high and explains more than 90% of the variance in the data for these relations. The results of Table 4 clearly show that a 1-unit change in X (time in years) corresponds to an expected value of Y (patents in transformers technology) by e^β = e^0.57 = 1.77, which is an expected increase of 77%. Mutatis mutandis for RNN a 1(year)-unit change in X corresponds to an expected increase in patents of e^β = e^0.29 = 1.34, 34%, whereas for LSTM, the expected increase in patents for 1-year change is e^β = e^0.33 = 1.39: 39%. Hence, the expected temporal increase in patents of transformer technologies is almost double compared to other alternative technologies.

Table 5. Parametric estimates of the model of technological evolution using data of new and previous (alternative) technologies.

Dependent Variable: Patents Concerning Different New Technologies Y
T = 10 Years	Coeff. b’1	Stand. Beta	Constant a’	F	R2	Period
New, Log LSTM10 years (Y)	1.07 ***	1.00	−1.23 ***	21671.5 ***	0.99	(2014–2024)
Previous, Log RNN10 years Explanatory variable (X)
New, Log Transformer10 years (Y)	0.94 ***	0.99	−0.88 *	309.00 ***	0.97	(2014–2024)
Previous, Log RNN10 years Explanatory variable (X)
New, Log Transformer10 years (Y)	0.87 ***	0.98	−0.24	213.31 ***	0.96	(2014–2024)
Previous, Log LSTM10 years Explanatory variable (X)
T = 5 years	Coeff. b’1	Stand. Beta	Constant a’	F	R2	Period
New, Log Transformer5 years (Y)	1.81 **	0.96	−8.36 *	50.72 **	0.93	(2020–2024)
Previous, Log RNN5 years Explanatory variable (X)
New, Log Transformer5 years (Y)	1.60 **	0.97	−5.57 *	56.62 **	0.93	(2020–2024)
Previous, Log LSTM5 years Explanatory variable (X)

Note: Y = patents of new technology (log scale, dependent variable); X = patents of previous technology (log scale, explanatory variable); and R² is the coefficient of determination. F is the ratio of the variance explained by the model to the unexplained variance. *** significant at 1‰; ** significant at 1%; and * significant at 5%.

shows the estimated relationships of the evolutionary model of technological substitution. Regression analyses over a period of five years (2020–2024) show that the log–log model estimation provides robust statistical results. The F value is significant (p-value < 0.05) and the coefficient R² is high and explains about 93% of the variance in the data for the transformer and LSTM technologies. The results of the log–log model estimation clearly show that a 1% increase in the patents of RNN increases the level of patents in transformer technology by 1.8%, whereas for LSTM it is 1.6%. The results show B > 1: a disproportionate and accelerated growth of patents generating a technological substitution of the transformer for RNN and LSTM (Table 5).

4.2. Pattens of Technological Invasion

Table 6. Parametric estimates of the model of technological invasion using data of new technology (predator–invader) and overall technological space in alternative technologies.

T = 10 Years	Coeff. b’1	Stand. Beta	Constant a’	F	R2	Period
Log RNN10 years	1.30 ***	0.99	−4.49 ***	824.94 ***	0.99	(2014–2024)
Log D10 years (Technological space—RNN) Explanatory variable (D’)
Log LSTM10 years	1.32 ***	0.99	−5.41 ***	1444.29 ***	0.99	(2014–2024)
Log D10 years (Technological space—LSTM) Explanatory variable (D’’)
Log Transformer10 years	1.19 ***	0.98	−4.95	191.16 ***	0.96	(2014–2024)
Log D10 years (Technological space—Trasf.) Explanatory variable (D’’’)
T = 5 years	Coeff. b’1	Stand. Beta	Constant a’	F	R2	Period
Log RNN5 years	1.22 ***	0.99	−3.67 ***	786.29 ***	0.99	(2020–2024)
Log D10 years (Technological space—RNN) Explanatory variable (D’)
Log LSTM10 years	1.39 ***	0.99	−6.06 ***	821.99 **	0.99	(2020–2024)
Log D10 years (Technological space—LSTM) Explanatory variable (D’’)
Log Transformer10 years	2.20 **	0.93	−15.23 *	26.98 **	0.87	(2020–2024)
Log D10 years (Technological space—Trasf.) Explanatory variable (D’’’)

Note: Y = patents of new predator–invader technology (log scale, dependent variable); D = technological space without predator–invader technology (log scale, explanatory variable); and R² is the coefficient of determination. F is the ratio of the variance explained by the model to the unexplained variance. *** significant at 1‰; ** significant at 1%; and * significant at 5%.

shows parametric estimates of the model of technological invasion using data on new technology (predator–invader) and the overall space of alternative technologies. Regression analyses over a period of five years (2020–2024) show that the log–log estimation model provides robust statistical results. The F value equal is significant and the coefficient R² is high and explains more than 85% of the variance in the data. The coefficient of regression in this model is a measure of technological invasiveness that for these new technologies is I > 1, indicating a high rate of invasion in the technological space of alternative technologies. However, the coefficient of technological invasiveness for transformer technology is 2.2, which is higher than RNN (1.22) and LSTM technologies (1.39), suggesting the highest power of invasion of the new transformer technology and of pervasiveness in the technological space of neural network architectures.

5. Analysis of Findings

Technology analysis of invasive behaviour in technologies that generate paradigm shifts in a short period of time provides critical information to explain scientific and technological development directed to the progress of human society [84].

The emergence of transformer technology is due to the interaction and convergence of competencies from mathematics and model design in neural networks [85,86,87]. Transformer architecture was introduced in the context of natural language processing, revolutionizing it, but it has been shown to be a flexible and powerful technology, finding new applications in diverse fields, such as computer vision, speech recognition, etc. [88]. The speed of the invasive technologies in transformers is a fundamental parameter to predict their ability to invade the scientific domain of alternative technologies in order to be a dominant one in the short run and drive technological, industrial, and economic change [89]. In fact, temporal and spatial models of technological evolution here, based on patent data, reveal the highest short-run rate of growth in invasive technologies of transformers (2020–2024) compared to other technologies. A basic driver of invasive behaviour in transformers is the interaction with different research fields and inter-related technologies [10,35,36,63,64,90,91,92,93,94,95,96,97,98,99]. Scholars have shown that technological interaction can support technological evolution, and this result is consistent with the multi-mode interaction approach by Utterback et al. [27]. In the case of transformers, technological interaction is generating high growth rates and a symbiotic-dependent evolution in which each technology benefits from the activity of the other inter-related technologies [10,93]. In particular, the technological interaction of transformers with other technologies generates synergistic combinations and fosters major innovations in different fields, opening new technological opportunities such as in healthcare (e.g., in ophthalmology, epidemiology, neurology, etc. [90,99,100,101,102,103,104,105]).

Moreover, transformers have invasive behaviour because they have the characteristics of a general-purpose technology, in short GPT [95,106,107]. Lipsey et al. ([108], p.43) define a GPT as “a technology that initially has much scope for improvement and eventually comes to be widely used, to have many users and to have many Hicksian and technological complementarities” [109]. Invasive technologies, having a characteristic of GPTs, exert a pervasive impact across industries and permeate the overall economic system. Bresnahan and Trajtenberg ([110], pp.86–87) show that GPTs radiate towards every industry and sector. In fact, transformers, having the behaviour of GPTs, generate clusters of innovations in several industries because they provide basic processes/components/technical systems for the structure of various families of products that are made quite differently, supporting co-evolutionary pathways, such as in autonomous driving [99], image detection with high-resolution remote sensing [90], etc. The manifold applications of transformers are driven by leading firms (such as Open AI, Microsoft, Google, Apple, etc.) to maximize profit and/or to exploit the position of a (temporary) monopoly and/or competitive advantage in specific industries [33,34,40,111,112]. In general, transformers are invasive technologies having the characteristics of both disruptive technologies and general-purpose technologies directed to “pervasiveness, inherent potential for technical improvements and ‘innovational complementarities’, giving rise to increasing returns-to-scale” (Bresnahan and Trajtenberg [110], p.83; Jovanovic and Rousseau [113], p.1185). Other characteristics of invasive technologies in line with the behaviour of GPTs are the scope for improvement, wide variety and range of uses, and strong complementarities with existing and potential new technologies [108,109,114,115,116]. Overall, then, the invasive behaviour of new technologies (e.g., transformer technologies) supports product/process innovations in several sectors for corporate, industrial, economic, and social change (cf.,

Table 7. Differences between disruptive and invasive technologies.

	Disruptive Technologies	Invasive Technologies
− Technology type	Radical technologies	General purpose technologies
− Technological dynamics	Destruction	Destruction and knowledge creation
− Technical characteristic	Pervasiveness and cost reduction	Pervasiveness and innovation spawning
− Business strategy	Exploitation	Exploration and exploitation (ambidexterity)
− Evolutionary patterns	Mutualistic interaction	Symbiotic interaction
− Rate of growth	High	Very high
− Period of diffusion	Medium run	Short run
− Industrial impact	Specific industries	All economic sectors
− Technological impact	Some innovations	Clusters of innovations
− Current Example	5G technology	Generative Pretraining Transformers

; [107,111,112,114,115,116]).

Invasive technologies are usually intentional, driven by new inventive pathways and the desire to improve efficiency of current technologies with consequential impact on dynamic capabilities of firms [19]. Technological invasiveness is characterized by the following properties:

• Rapid spread: Invasive technologies rapidly grow, overwhelming existing technological systems. In particular, the behaviour of invasiveness in new technologies can spread rapidly across sectors and economic systems, often driven by consumer demand and other market forces, new human needs, and strategic goals of organizations, including nations. Conversational AI assistants, for instance, have quickly become ubiquitous.
• Disruption of existing ecosystems: Technological invasions disrupt technological, social, and economic systems, changing how people and firms interact, behave, and conduct business. Technological invasiveness (e.g., smartphones, AI technologies, etc.) can significantly alter communication systems, social interactions, and even business models. New invasive technologies render established technologies obsolete, leading to significant changes in industries and job markets. The rise of digital photography, for example, has largely replaced traditional film photography in different industries [117,118].
• Adaptability and evolution: Invasive technologies are flexible and adaptable technologies, tending to spread more easily in various sectors. Advancements in invasive technologies occur rapidly by integrating with inter-related technological systems that generate clusters of radical and incremental innovations. Successful technologies that have the behaviour of invasiveness often adapt and support other inter-related technologies, such as cloud computing, which has numerous applications in industries, from healthcare to finance [119]. Generative AI intelligence is used in healthcare, smartphones, laptops, new sensor technologies [119,120,121], quantum technologies [85,86,122,123], etc.
• High impact: The primary impact of technological invasiveness is on human systems, such as economies, industries, and social structures, and then the impact can extend to the environment and other systems. Invasive technologies generate high economic impact with job displacement, market development over time and space, etc., that change products and processes with new functions that improve their scope to solve problems and satisfy needs in society. In short, invasive technologies impact human societies, economies, and cultures, influencing how people live, work, and interact.

However, invasive technologies have both positive and negative impacts. Although they offer benefits like increased efficiency and connectivity in products and processes, organizations, etc., invasive technologies can also pose risks such as privacy concerns, security, health and job displacement, etc. [119,124,125,126]. New invasive technologies may lack regulatory frameworks or ethical guidelines, leading to unchecked growth and potential misuse that deserve appropriate measures and organizational behaviour [127]. Hence, new invasive technologies need ethical and legal frameworks designed to properly manage their deployment and impact on economies and societies [128].

6. Concluding Remarks

Advances in information sciences are generating new technologies that introduce significant changes to economies and societies. This study proposes, for the first time, the concept of the invasive behaviour in new technologies. Successful technological invaders can have a devastating impact on economic and industrial systems and overall human society. The proposed theoretical framework of invasive technologies can clarify the main characteristics of ongoing technological change for supporting R&D management implications and innovation policies to guide emerging technologies with a high potential effect in almost every sphere of human activity in the current information, digital, and Artificial Intelligence (AI) era [129]. This study verifies the theory of invading technologies by focusing on transformer technologies that have unparalleled growth at the expense of other alternative technologies, creating the basic conditions of a technological revolution directed to generate a drastic technological change with clusters of radical innovations that have significant effects on economic and social systems in a not-too-distant-future. Invasive technologies have the behaviours of rapid diffusion, destruction of other technologies, and seizure of their scientific, technological, and commercial spaces. To put it differently, invasive technologies are both disruptive and general-purpose technologies that make obsolete established products and competencies with a high pervasiveness in manifold industries over the short run with long-run impacts [29,30,31,36]. The invasive dynamics are based on competition for higher performance and the effectiveness of new technology in problem-solving activities [26]. What this study contributes is that the invasive behaviour of new technology is more drastic than disruptive technology because it has also main characteristics of general-purpose technologies being highly inter-related with manifold technologies and applied in different sectors, such as transformer architecture [130]. In this context, the rapid evolution of invasive technology paves the way for the development of other technologies by “expanding the adjacent possible” [16].

6.1. Theoretical Implications

The predictions of our theoretical framework in invasive technologies are borne out in the phenomena investigated in Artificial Intelligence technologies, paving the way to a better understanding of innovation processes in a knowledge economy.

The basic characteristics of invasive technologies are as follows:

• Rapid diffusion and acceleration, outcompeting the growth of alternative technologies.
• Pervasiveness over time and space in the short run.
• High interaction with manifold technologies, generating symbiotic growth.
• Generalist behaviour and adaptation: invasive technologies adapt to a variety of structures generating new and improved products and processes.
• Competitive advantage: invasive technologies have a competitive edge over alternative technologies, with rapid growth and/or efficient resource utilization.
• Disruption of previous technologies and creation of new ecosystems: invasive technologies capture the scientific and technological space of other technologies. These new technologies also change dynamic capabilities (the organization’s ability to integrate, build, and reconfigure internal and external competencies to address rapidly changing environments [19]).
• Economic and social impact: invasive technologies cause significant economic benefits by affecting different industries and supporting social change.

The results here can be the basis for an emerging science of invasive technologies that explains technological, economic, and social change considering the following aspects:

• Invasiveness of new path-breaking technologies;
• Invasibility of innovation ecosystems and the creation of new ones;
• Interaction (patterns of the new technologies) × (ecosystem interactions) may support technological invasion based on a set of concurrent aspects and an identifiable pattern in different industries [131].

6.2. Managerial and Policy Implications

Invasive technologies tend to have similar patterns based on two contrasting forces that can have managerial implications: The tendency to retrace already explored avenues (exploit) and the inclination to explore new technological opportunities. Policymakers and R&D managers can use the rate of growth in invasive technologies to make efficient decisions regarding the sponsoring of specific technologies that have a high rate of growth (invasion) to foster technology transfer for boosting industrial change. Managerial approaches in invasive technologies can be underpinned in the framework of the expansion of the adjacent possible, in which the restructuring of the space of technological, economic, and social possibilities is conditional upon the occurrence of radical innovations.

Proposed theory and empirical findings can guide an ambidexterity strategy of innovation management for invasive technologies by balancing exploration and exploitation approaches directed to adaptation in turbulent environments to achieve and sustain competitive advantage [132,133,134].

Nations and governments can support public R&D investments in invasive technologies through grants and incentives in order to address societal or strategic challenges. For instance, R&D investments in the Flexible Innovative Transformer Technologies to develop advanced transformers that can provide real-time monitoring of voltage and temperature in order to enhance grid resilience and address supply chain constraints [135]. These new transformers are designed to be flexible, modular, and scalable, making them suitable for a range of different applications from distribution to transmission. They also incorporate advanced materials and designs to improve performance and reduce costs [135].

General managerial and policy implications to support the integration of new invasive technologies into society and the economic system can be based on the following strategies:

• Regulation and policy development: Governments and regulatory bodies can develop standards and guidelines to ensure that invasive technologies are safe, ethical, beneficial, and used responsibly. Establishing ethical frameworks in the emerging phase can guide the development and deployment of new invasive technologies to align them with societal values and not cause harm. In particular, involving diverse stakeholders of the ecosystem in these new technologies can improve decision-making processes, help address ethical concerns, and ensure technologies serve the broader community.
• Education policies about new invasive technologies can help foster informed adoption and use. In addition, training for professionals and users can promote understanding and effective utilization of new invasive technologies for their positive impact in specific contexts, such as in the health and education sectors.
• Investments in infrastructures can be developed to support new innovation ecosystems directed to support new invasive technologies and related integration, such as in digital pathology, etc.
• Public–private partnerships can leverage expertise and resources to support the development and responsible integration of new invasive technologies. In addition, international cooperation in strategic fields can address global challenges and ensure that new invasive technologies have effective and beneficial effects on a global scale.

In particular, the best practices for managing technological invasiveness in firms are as follows:

• R&D investments directed at innovation development, the adoption of new invasive technologies, and their adaptability to the rapid pace of technological change.
• Training programs to keep human resources updated on the latest technological advances and security practices.
• Comprehensive decision-making involving stakeholders, employees, customers, and partners, to understand new problems and needs for improving invasive technologies.
• Implementation of security measures in order to ensure that data are protected through encryption, firewalls, and regular security audits.
• Development of ethical guidelines for the use of new invasive technology within organizations.

Instead, innovation policies of nations for managing invasive technologies can be as follows:

• Infrastructure investments to build the necessary innovation ecosystem to support the adoption of new invasive technologies.
• Public R&D to drive innovation and address emerging challenges in society.
• Public education to train citizens about new invasive technologies and their potential impacts in practical contexts.
• Collaboration and partnerships between different subjects (government, industry, and academia) to leverage know-how and use of resources directed to invasive technologies. International collaboration to develop and implement new invasive technologies aimed at addressing global challenges by ensuring responsible use in practical contexts.
• Development of ethical frameworks to guide the proper use of new invasive technology in industrial, social, and economic systems.
• Design of comprehensive regulations that address the ethical, security, and privacy implications of new invasive technologies combined with international cooperation with other nations to develop common and appropriate regulations.

6.3. Limitations and Development of Future Research

This study shows for the first time, to the best of our knowledge, the behaviour of invasive technologies to explain some new aspects of technological and social change in knowledge economies. However, these conclusions are, of course, tentative. This study provides some interesting but preliminary results in these complex fields of emerging technologies, but some limitations to consider in future studies can be summarized as follows. Many fundamental questions in the science of invasive technologies can only be answered through integrative studies, such as interdisciplinary research that encompasses comprehensive studies of invasive technology in a specific field, and comparative studies of invasive behaviour of the same technologies across multiple fields and industries. In this study, the invasive behaviour of technology focuses on a field dominated by a single dominant invader–predator technology (transformer). However, studies of multiple invasive technologies are mostly lacking. Such studies are, however, important to understand the dynamics of new paradigm shifts generated by invading technologies, also with interactions among multiple technological invaders. In the context of invaded technological ecosystems, an emerging challenge is also to understand the role of gradual changes in technological and environmental factors in determining invasion trajectories over time and space [136]. Hence, it is interesting to compare the invasive behaviour of the same technologies across multiple industries and research fields, to explain how “invasiveness” affects different technological ecosystems. Analogous to biology, the impacts of invasive technologies are strongly co-shaped by technologies and environment interactions [137,138], which can only be understood through comparative studies across different technologies and industries [131]. More studies that compare the behaviour of technology in native research fields and the invaded spaces of inter-related technologies are needed to explain the macro-evolution and impact of new technologies with technological mutation and adaptation to different contexts in socioeconomic systems [139,140].

These studies are needed in the future because the investigation of only one technology is very likely to arrive at spurious conclusions [141,142]. For instance, characteristics that are most frequent among invasive technology in a specific market might not be relevant for predicting the behaviour in other industries and fields. In fact, a future idea is to verify if the superiority and flexibility of invasive technologies are aspects that apply to all technological spaces in different industries [142,143,144,145,146].

Other limitations are that patent analyses can only detect certain aspects of the ongoing dynamics of invasive technologies and the next study should apply complementary analysis with scientific outputs considering confounding factors (e.g., level of public and private R&D investments, international collaboration, etc.) that affect the evolution of new invasive technologies over time and space.

In short, there is a need for much more detailed research into the investigation of invasive technologies in socioeconomic systems to clarify evolutionary patterns for technological, economic and social change. Despite these limitations, the results here illustrate that invasive technologies can clarify basic characteristics of current technological change. This study is a starting point and encourages further theoretical exploration in the terra incognita of invasive technologies within and between scientific and technological domains that have rapid change in the new era of Artificial Intelligence. These technology analyses are basic for improving the prediction and evolutionary pathways in emerging technologies and supporting R&D investments towards new technologies and innovations that have a high potential for growth, invasion and impact on socioeconomic systems. However, a comprehensive explanation of sources and diffusion of invasive technologies is a difficult topic for manifold complex and inter-related factors involved in science and society, such that Wright ([147], p. 1562) properly claims that “In the world of technological change, bounded, rationality is the rule”.