Generative Artificial Intelligence – New Horizons for Technology Management?

Since the release of OpenAI’s chatbot ChatGPT in November 2022, records for the development and use of tools and programs based on artificial intelligence (AI) have been broken almost every month. One million users were registered after just five days. ChatGPT reached the important 100 million user mark after two months. The previous record holder for the fastest user growth was TikTok after nine months. This made ChatGPT the fastest growing consumer app in history [1].

Ever since this high-profile launch, both private users and companies have been looking for the correct way to use similar software platforms in light of their potential risks. For example, the Samsung Group banned its employees from using AI tools such as ChatGPT after it was discovered that employees had uploaded sensitive code to the platform [2]. However, the potential for increasing productivity is also evident – a study by scientists at MIT investigated the impact of using ChatGPT for the creation of business documents and found that ChatGPT significantly increased productivity: the average processing time for text documents fell by 40%. At the same time, the quality of the results increased by 18% [3].

Enabling utilization of generative AI in an industrial context

The use of generative AI when integrated into easy-to-use applications that are user-controlled with natural language promises a significant increase in productivity in activities that have historically been assigned to the domain of “human intelligence”. It is conceivable that conceptually demanding and even creative activities could be automated – thus extending automation well beyond the shop floor and routine manual processes.

Such activities can also be found in technology management. Technology management (TM) deals with the critical task of ensuring that the right production, product and material technologies are available to a company at the right time and at an acceptable price point. TM therefore has a considerable influence on the competitiveness of industrial companies [4].

Technology management is particularly important in times of highly dynamic technological development and shortening product life cycles. However, many of the tasks involved in technology management are time-consuming and resource-intensive, such as literature, patent, or trend research or the preparation and communication of technological information as a basis for decisions.

Considering the worsening shortage of skilled workers and the ever-increasing, almost inexhaustible amount of potentially relevant information, the question arises: Can technology management activities be supported or even completely automated by generative AI? To examine this question, not only for the current technological status quo, but also for the potential of generative AI in the future, various development horizons, i.e. conceivable future development stages, are considered.

Development horizons of generative AI

Horizon 1: Specialized generative AI

Specialized generative AI describes the current state of the art in generative AI models. These are focused, highly specialized models that have been designed and trained for dedicated use cases. This includes, for example, the generation of images based on short text inputs (prompts) by Dall-E or the evaluation of extensive text data by ChatGPT.

These functions might appear comprehensive and multifaceted, yet in terms of generalization capabilities and breadth of applicability they exhibit at best isolated and limited characteristics of the subsequent horizons [5]. Corresponding solutions are therefore only suitable for the intended closed applications and require that corresponding activities can be strongly structured and tailored to the suitability of the models.

Horizon 2: Integrated AI

This horizon is less dedicated to describing the progress of the individual models, but rather provides a cross-model architecture that meaningfully integrates various specialized generative AI solutions and also networks them with human intelligence in a targeted manner [6]. This also opens up semi-structured applications, which may have to be run through in a series of steps and which successively build on different models in varying combinations [6]. A simplified example of this would be the integration of a chatbot into the creation of an order sketch and the automatic checking and dispatch of the finalized order.

This horizon also envisions the increased integration of AI into the structures of companies, which would increasingly enable the automated use of company and case-specific context data [7]. Integrated generative AI cannot be generalized at will or used flexibly across the board, but its integration partially reduces the limitations associated with the specialization of individual models in Horizon 1.

Horizon 3: Generalized AI

The third horizon, which has long been the target of numerous AI research efforts, does not merely enable more flexible use through integration, as in Horizon 2, but rather creates fundamentally more far-reaching models [8]. These models achieve a significantly greater breadth of applicability and generalization capabilities. They are thus able to carry out a wide range of activities and achieve goals in varying contexts and environments – even in ways that were not necessarily anticipated during modeling.

These models differ especially from the other horizons in that they are much less application-specific – the models can be applied to problems for which the AI was not specifically trained or designed, and even to ones that were not considered at all during development [8]. This is accompanied by an increasingly human-like ability to make judgments that can take into account a wide variety of structured and unstructured data on companies and their respective fields. Complex knowledge transfers will also be made possible.

In order to identify feasibly supportable or automatizable activities for the various development horizons, a comparison was made between the potentially eligible activities of technology management (TM) and the development horizons of generative AI. For this purpose, TM activities must be described and characterized with regard to their “AI automatability”.

The information and data basis that is integrated into the activities (i.e. the foundation on which generative AI would ultimately operate) and the predominant type of activities with regard to the capabilities of the various horizons are particularly decisive factors in the characterization. Essentially, TM tasks and functions can be described as falling into one of four categories (Fig. 2).

Technology management activities as a starting point for AI-supported automation

Orientation and interpretation

This includes primarily technological foresight tasks, such as scanning, scouting and monitoring, which are characterized by research and the evaluation of information [4]. In particular, unfocused scanning to identify new technologies or solution approaches outside of the well-known and comparatively narrow scope of one’s own company requires the time-consuming collection and processing of information from sources outside the company. However, a more targeted aggregation of information , e.g. to describe specific technology alternatives, also requires the consideration of content that is limited in focus but extensive in depth.

This information, such as patents, specialist literature, or product brochures, is comparatively highly structured. The processing of information required for the final orientation and interpretation includes the transfer of individual data or information into context-related knowledge, for example the maturity and feasibility of a technology as assessed for an investigated use case or the assessment of a technology’s advantages and disadvantages. Relevant information , such as the specialist knowledge of technology experts, is frequently available in an unstructured form [9].

Positioning and decision-making

Based on the previously conducted orientation and interpretation activities, the collected information is analyzed and then transferred into concrete decisions and the strategic orientation of the company (positioning). The core component here is the strategic positioning of the company in individual fields of technology, which is commonly determined in the dimensions of performance, timing, and technology source. The competencies and resources currently available play a decisive role in determining which position is ultimately sought in the competitive environment, given specific technological possibilities and environmental factors.

Particularly in volatile fields, it is important to bring together and analyze both external and – increasingly – internal company information in a structured manner (company and environment analysis) [9]. The focus in this field of activity is increasingly shifting to complex, highly context-dependent linking of information, as well as the – occasionally – creative transfer to decisions. Today, companies incorporate extensive empirical knowledge and intuition, information that only has a low degree of structure, into these decisions.

Organization and planning

Once decisions have been made and strategic positions have been selected, they are operationalized and then implemented. Corresponding measures, projects, and resources thus need to be planned and organized. These activities primarily involve company-related internal information. Structured processing and communication within the organization are essential at this stage.

Implementation and usage

With the switch to the concrete implementation of the previously defined plans and measures, supervisory and monitoring activities come to the fore. These include the communication of objectives, the continuous comparison of planning with actual developments, or operational project management. The relevant information for this stage is almost exclusively concentrated in sources within the company.

Exemplary case studies for the use of generative AI in technology management

Case study 1: Partial automation of orientation and interpretation

In cooperation with Hilti AG, two exemplary use cases were used to test the extent to which potential can be harnessed using generative AI: In order to investigate the extent to which generative AI can be used to automate the process of orientation and interpretation in technology management, it is first necessary to define which tasks occur. The four focal areas of orientation and interpretation are: definition of technology fields, technology scouting, technology field evaluation and subsequent technology monitoring [4].

The approach developed to enable the automation of orientation and interpretation in technology management was tested in three separate sub-processes. The technology search could be partially automated via targeted queries to a generative AI by querying sample technologies in a technology field (heat treatment) and requesting output of potentially relevant production technologies in this area. In addition, the evaluation and classification of generative AI was also supported. Initial classifications and evaluations were carried out by specifying the topic, type, usage scenario, and technology maturity level of generative AI for the previously identified technologies.

The results of these two work steps could then be combined to form a usable data basis for an automatically generated technology radar to visualize the results (Fig. 3). In this context, it was shown that generative AI can automate the search, evaluation, and classification of technology with the right instructions. This requires establishing diverse and specific sub-steps as well as delivering precise instructions to the respective generative AI.

However, there are also various risks associated with the use of generative AI in this use case. For example, an outdated data basis can lead to outdated technology outputs or assessments. The lack of insight into the model’s conclusions when processing prompts leads to a lack of traceability in the specified solutions. In addition, it cannot be guaranteed that the statements made by generative AI models are true (the tendency of current AI models to invent false statements is also referred to as hallucinating [10]).

For this use case, using generative AI models has proven to be quite beneficial, as it can greatly reduce manual effort in early technology identification and evaluation. Self-written scripts can be used to link individual applications and provide promising information.

Case study 2: Creative and combinatory assistance in the strategy process

The second case study in collaboration with Hilti AG examined to what extent the strategy process could be made more efficient by the incorporation of creative and combinatory assistance. In the strategy process, the description of a company’s technological resources and capabilities is first used to generate a competence formulation, and an environment analysis is carried out to determine potentially relevant areas in which the company could actively develop or innovate products and services in the future. The formulation of competencies and the determination of the field of innovation are then used to formulate a vision, on the basis of which strategic positioning can take place.

Large language models, i.e. very large AI models that have been trained on huge amounts of data and understand the relationships between words and phrases within such datasets, can be used in this context to generate suggestions for competence assumptions in a competence analysis. Furthermore, creative support could be implemented in the strategy process, e.g. by simulating the Walt Disney method as a creativity technique to create a technological vision of the technology and innovation fields [11].

In the case study, an AI model prepared for the role of the Walt Disney method was integrated into strategic considerations on a test basis, resulting in an often unexpected but nonetheless content-enriching perspective and a fundamental acceleration of the strategy process. In addition, communication templates suitable for the target group were generated directly afterwards.

If translated into a general approach, such a use case is associated with various risks. For example, the respective results – and therefore also their success– are significantly influenced by the training data that was selected. In addition, depending on the utilized generative AI model, the confidentiality of sensitive information cannot be fully guaranteed. Generated impulses should hence always be scrutinized, especially for strategic decisions.

In the process implemented here, it was found that AI-based suggestions certainly serve to stimulate the creative process. The search for links between different content worked reliably for the most part and adapting the communication to the target audience was comparatively straightforward.

In summary, the example case studies used to test the first development horizon of generative AI could show that generative AI can already be used in many areas of technology management with the currently available technological maturity. However, these applications are not yet particularly focused and require precise and incremental instructions.

The creation of effective links between individual work steps with the support by generative AI is already possible with simple programs. An understanding of what constitutes adequate instructions is therefore a key success factor for almost any application scenario in technology management. Both case studies are limited to the applications tested to date in the first development horizon, which the team of authors has been investigating since the breakthrough of generative AI.

For the second development horizon, for which the first applications are already commercially available at the time of publication, it can be observed that integrated generative AI enables the automatic execution of entire process chains. In this way, companies can react much faster to changes in their environment. In addition, generative AI can significantly relieve the burden on human workers by accessing a company’s digital knowledge base. A well-maintained database is an essential prerequisite for this.

The road ahead

In summary, it can be said that the use of AI can facilitate or significantly accelerate various processes in technology management. Regarding use cases of the first development horizon, this means an acceleration of similar processes that were previously carried out manually. For use cases in the second horizon, existing processes can potentially be redesigned and, regarding the third development horizon, a potential redesign of the entire previous way of working in technology management can be achieved.

It can be concluded that the processing of both operational and strategic tasks in current technology management will change as a result of the increasing adoption of generative AI. Systematic technology management will be possible with fewer resources and the use of generative AI will result in a knowledge advantage for the companies that adopt generative AI in a targeted manner and are thus set to gain a competitive advantage in the long term. Overall, five theses on the new reality of generative AI can be derived from the findings of the conducted case studies:

1. AI is only as good as its data: the digitalization and standardization of existing knowledge significantly influences the success of projects for the application development of generative AI.
2. Prompts are the new language: in order to successfully apply generative AI in your own company, employees need to be trained in how to interact with such models. Prompt engineering, i.e. the targeted formulation of instructions for generative AI applications, is essential for the quality of output results, at least in the medium term.
3. From musician to conductor: The tasks performed by human workers will change significantly with the spread of generative AI. In order to successfully deal with these new tasks, the structuring and delegation skills of employees must be fostered.
4. Generative AI, the new coworker: With the constant development of generative AI, its contribution will become more and more relevant for companies and its potential will increase. To keep pace with this development, companies should start decentralizing decisions as early as possible.
5. The courage to leave gaps: Not every generative AI application is useful or effective. In order to exploit the full potential of generative AI application scenarios, companies should test various applications and then anchor them in their own strategy.

This study was funded by the European Commission’s H2020 project EPIC under the grant number 739592.