IEEE Future Directions
Related Articles
Let me spend a few more words on this “AI in the small” from the point of view of the evolution of AI (both as a consequence and as a fuel to its evolution).
As shown in the graphic (bear with me, I am not a professional designer and I didn’t ask AI to create the diagram) there are a few big players that have invested, and are investing, money and resources in the training of transformers (by accruing billions and billions of data). These players offer API -application programming interface- (and web interfaces) that can be used by third parties, companies and individuals, to generate value (Commmercial Applications – on the top right hand of the stack).
Those same APIs can also be used by a company to interface with a Provate Specialised Model that contains, and is continuously being expnaded, refined, through the data that company is generating and harvesting in its operation. As indicated in the graphic these data derive from:
- IoT generated data: the pervasive presence of IoT both in the supply chain, shop-floor, distribution chain and in the products themselves (including virtual IoT embedded in services and software packages) keeps generating data that are both significant in volume (they pile up over time)and in patterns (the way they are arriving generating a variety of correlation output). Most of these data, today, are used locally -like to control a robot- and then they are discarded. The value that can be derived from their correlation and the knowledge that can be derived from inserting the correlations in a model is not leveraged.
- Data Processing at the Edge: clusters of IoTs as well as devices, equipment connected in a local area network can engage in a variety of interactions. These can be recorded and processed, along with other data generated by IoT, generating further data that express the “meaning” of what is going on. All these data resulting from processing at the edge can be used to enrich the model and power local intelligence.
- Internal Knowledge: any company has a rich set of internal knowledge, how it operates, the resources managed (including workers) and the way they are orchestrated, the relations it has with the various players in the supply chain, the ones with their customers, the vision and understanding of the market … plus the processes being used. In the end what characterise a company is the set of processes through which it operates. There are several ways of formalising this internal knowledge and to keep it up to date. One way is through the use of Cognitive Digtal Twins.
- All data, and interactions, resulting from the use of the company created intelligence, i.e. how it is being used, what is found appropriate and what is leading to further refinement, be it from company resources and from the end users, in case these are provided access to the company intelligence -an interesting proposition that can result in service offering!- can also become part, lead to a refinement of, the Private Specialised Model, as indicated by the green dashed line (this flow of data is indicated differently from the previous one because it is an integral part of the artificial intelligence feature, leading to a continuous learning and improvement). Notice that I have also indicated a similar process in the Commercial Applications (red dotted arrow). This is, by the way, one of the issues that have been raised by those opposing -or just criticising- the new wave of generative AI, the potential loss of privacy. The capturing of the interactions and data implied in its use is clearly revealing aspects of the user (this is no different to the use of a search engine where it learns what you are searching and therefore deduce information about you!). OpenAI is stating in its accompanying information on ChatGPT use that prompt strings submitted via API are not recorded whilst the ones submitted through a form (that is most likely what you are doing in using ChatGPT, unless you are a programmer and have developed an app to access it -and payed for the access-) can be used to improve the algorithm (a subtle way to tell you that they can do whatever they like with your prompts and with the answers generated).
This knowledge, along with all the data created and harvested, can become an integral part of the Enterprise model forming the Private Specialised Model indicated in the graphic.
There is an enormous value in this possibility of flanking the LLM provided by the big guns through the Commercial API with a Private Specialised Model. The first feedback coming from industry (we are just starting to see this happening) indicate a gain in productivity of the order of 5 to 15%.
That’s a lot. It can transform a company into a leader in a given market sector. particularly so for those companies operating in consolidated market where the winning factor is competition on price.
To get a feeling on how this can be done read this article.
