AI Hallucination & Professional Transformation
Pretext
In 2021, together with a seed investor, we founded the startup iMBdigital.Gallery_. iMBdigital.Gallery_ was founded with a focus on NFT, Blockchain and Web3 technology. The initial idea was born from a combination of love for photography and high interest in blockchain-based digitalization. From the beginning, image-based AI (artificial intelligence) and ML (machine learning) was an essential part of the startup.
In the meantime, the startup is active in some B2B projects and support in projects of iMB. For example, in AI and ML-based process mining and feasible applications with blockchain tech. This always revolves around the analysis and determination of event logs and the corresponding processes, the analog and digital documents and their organization, archiving and channels of forwarding. Especially in view of the document analysis in connection with the corresponding event log at any time in the value chain, ML and consequently AI is a fundamentally important analytical colleague when it comes to plausibility checks. The productivity gain and the level of detail increases exponentially.
Obvious impacts
From all observations, we are sure that an epochal change e.x. in the whole process of so-called BPO is in the offing. Also, entire administrative processes will be changed so massively that actually no brick will remain on the other.
However, it is also important to remain critical. The results of ML and AI cannot, or rather should not, be adopted without reflection.
It is becoming apparent that the professionals who work with ML and AI systems must have a very profound knowledge of processes in order to subject process mining to a reverse plausibility test. In addition to the knowledge of processes, a fair amount of general knowledge is required to program the AI in prompt engineering with NLP (Natural Language Processing) in such a way that the desired queries also find their way to the result. However, this also applies to the phase of interpreting the results. The "technical idiot" can cause much more damage here than in the previous honest analog system, where a linear Excel-based verification tends to take place.
Sounds scary - but it's not!
What we are talking about here is the so-called AI hallucination. We have had broad experience with this in a wide variety of contexts. And it goes on almost daily. It is also always a matter of knowing when you can use AI hallucination for e.g. Critical Hypothesis Testing in a meaningful way and when the digital assistant outputs supposedly correct results but they do not stand up to a reverse plausibility check.
Let's dive a little deeper!
In the following, we want to discuss various aspects of AI Hallucination in this blog post based on our project experience. We have repeatedly discussed all of the aspects mentioned with our clients as well as internally at iMB and iMBdigital.Gallery_. If you follow us on the various social media channels, you will have found similar discussion posts fragmental again and again.
What is AI hallucination?
AI hallucination is a technique in which an AI system generates output that does not exist in the input data. It is sometimes referred to as “imagination” or “dreaming”, as the AI system is able to create novel objects or scenarios that are not present in the input data. AI hallucination can be used to generate creative and unique images, or to produce anomalous results (such as a face where no face should be). Very well known examples can be found quickly by anyone or generated by themselves using e.x. ChatGPT, DALL-E or MidJourney.
Which are the dangerous of AI hallucination?
AI hallucination can be dangerous if the output created by the AI system is overly realistic, or if the output contains sensitive information that could be used for malicious purposes. Additionally, if the AI system is not properly supervised and monitored, it can create unpredictable and potentially dangerous results. For this reason, AI hallucination should only be used with appropriate safeguards and oversight.
Is it possible to use AI hallucination in a positive way?
Yes, AI hallucination can be used in a positive way. For example, it can be used to generate new ideas, images, or scenarios that may be useful to a particular project. Additionally, AI hallucination can be used to generate creative and unique images, or to produce anomalous results that can lead to deeper understanding of a particular topic.
Especially in the support for the creation of scenarios AI hallucination can and should be used consciously. In the old view of scenario creation, a massive mathematical model was mostly relied upon to design a generation. On the one hand, this cost a lot of time and was actually extremely limited, since setting up the model took a lot of time, the input had to be done by mostly Excel specialists, as well as the analysis and evaluation.
This process is of course not agile. Today, it is replaced by the participatory design thinking approach and then verified by LoFi model testing. In the design thinking phase, AI hallucination can and should be used consciously to accelerate the project group in the creative process.
What can be done to avoid AI hallucination?
AI hallucination can be avoided by using appropriate safeguards and oversight. This includes careful monitoring of the AI system, as well as making sure that the output generated by the AI system is appropriate and safe. Additionally, it is important to ensure that the AI system is properly trained and tested before it is used in any real-world applications.
The ability of contextual thinking of the AI user can avoid AI hallucination?
Yes, the ability of contextual thinking of the AI user can help avoid AI hallucination. This can help ensure that the output generated by the AI system is appropriate and safe. Additionally, it can help the AI user to better understand the context in which the AI system is operating, allowing for more informed decisions.
Which are the important points to develop a contextual thinking?
The important points to develop a contextual thinking are: understanding the environment in which the AI system is operating, being aware of the data and algorithms used by the AI system, and being able to interpret the results produced by the AI system. Additionally, it is important to be able to recognize potential errors or anomalies in the output generated by the AI system, and to be able to take corrective action.
Likewise, it is more than advisable to have a good and broad general education - the root, the core, to develop contextual thinking.
Often, the question arises more and more whether we will soon see completely new professions with AI.
Probably.
Professional transformation through AI
If we currently talk to specialized recruiters or startups, it quickly becomes clear that AI will quickly transform the job market and the demand for developers, analysts and prompt engineers is strongly fueled.
During the course of the pandemic, I had already taken advanced training to become a certified digital engineer at the HassoPlattner Institute, University of Potsdam, Germany. Already during this advanced training, the areas of specialization within my studies were Design Thinking, Process Mining and AI as well as ML in conjunction with Big Data Analysis. This study, together with the love for photography, had led me and colleagues to found the startup iMBdigital.Gallery_ at the end of 2021.
What are we currently observing in the Brazilian labor market with respect to AI and ML?
Basically, it has to be said that a fair number of professionals in the so-called white collar sectors (lawyers, accountants, consultants) have not yet grasped the revolution that is hurtling towards them at breakneck speed. So the enthusiasm is simply to use e.g. ChatGPT as a text generator ... . OK. …
Link: OpenAI
It is likely that the further course of digitalization will be massively influenced by ML and AI. At this point, it should be clear that these topics are not new. What is new is the speed with which the models can be trained and the application of NLP in prompt engineering. Thus, it should be absolutely assumed that the demand for professionals as well as the character of transformation projects will be increasingly determined by the focus on ML and AI.
Especially in 2023, we can clearly observe that there are already the first companies asking for pilot projects using ML and AI. Especially multinational companies are also evaluating small pilot projects in Brazil and Mexico here. It seems very clear that the application has already broken the boundaries of early entrants, startups and pure technology companies in its scope and impact. Without a doubt, this trend has already reached the so-called traditional industry.
In our conversations with clients, startups as well as recruiters, the following new jobs are mentioned again and again, which people seem to be looking for already, or which they will focus on soon:
AI Specialist: This is of course a very general description. But if I decipher the term correctly, it means a professional who is involved in the creation, data training and testing of AI systems. The requirements always revolve around the areas of programming languages, mathematics and statistics, prompt engineering algorithms and data structures, and the ability to curate data.
Chatbot developer: This is a professional who works in parallel with the AI specialist to generate the appropriate AI bots, plan the implementation, and set up support tools. This is the craftsman of prompt engineering. This professional also needs deep knowledge of programming as well as a sensitive understanding of the users' side in the chatbot's group of customers.
Chatbot technician: this is the maintenance technician of the chatbots. This professional should have very good knowledge in the field of ML applications, as this is where the data streams are directed into the analysis. Also, this professional must have a good operational understanding of NLP.
Chatbot Analyst: Parallel to the chatbot technician, this professional is probably the operator of the bot. He configures the chatbot, analyzes the data and emits the corresponding reports. Central is also the task to ensure a constant improvement with the technician. Application-oriented profound knowledge in programming and prompt engineering, analysis skills with focus on patterns are central skills.
UX Designer with focus on Chatbot: This professional designs the user and consumer experience within the AI Chatbot application. The ability to generate intuitive navegation is central to this role. Visual supports in the AI Chatbot function are extremely important. It is imperative that this professional has deep knowledge in design thinking, data analytics, and understanding the using customer.
Data Engineer: This professional is active in the development and maintenance of the AI Chatbot. The basis is the data, from the testing phase to the ongoing operation. The professional needs profound knowledge of ML applications. The data engineer is the central professional in the development of the training model of the AI chatbot, as well as in the constant adaptation and evolution of the response behavior of the chatbot. In this role, the professional is the "daily conscience" of the chatbot to fend off attacks, balance tendencies, and make AI hallucinations in the response behavior controllable and detectable. It needs to extract, filter, clean, and cart the data. Key skills are data modeling, API, and a high level of general knowledge.
AI Chatbot Manager: Managing an AI chatbot means having a high level of general knowledge and education. The responsibility is very high. He must be able to lead his team and moderate technical professional knowledge with general education. An increased knowledge in post-modern ML and AI technology is fundamental. Project management and design thinking and evaluation of LoFi models must be the daily business of this professional.
And what will the overall requirements be?
It's not quite clear yet. But anyone who deals with ML, AI and prompt engineering should in any case have a broad general education in order to be able to classify exactly what the limitation of each ML and AI means for the created prompts and how one can use creative AI hallucination. The general knowledge here helps balance limits and use the technology responsibly.
We ourselves had used the AI hallucination in the chatbot of ChatGPT of OpenAI quite actively since Q3 2022, in order to circumvent the limitation of the version of GPT3 at that time, but at the same time to properly prompt the result in the socio-historical context of Brazil from our point of view and also to correctly classify the results.
In this context, the issue of the threat to jobs and functions comes up again and again.
In general, it can be said that the higher the scalability and the possibility of data collection, the higher the substitution by AI.
From my own experience, three areas currently stand out.
Programmers
Programmers have always seen themselves as the leading edge of technological and digital development. If we take the chatbot ChatGPT4 as an example, we see that the bot has learned a total of four programming languages in its training phase through OpenAI. It seems almost ironic that the chatbot could now even substitute a large part of the profession of creators.
The revolution eats its children ... .
Of course, this does not mean that programmers will become obsolete! Quite the opposite! But the large mass of average programmers will probably no longer be needed. The top people will enter the high speed elevator to new challenges.
What will be noticeable is the strong spread in the payment of programmers.
Prompt Engineers
Yes, that's right, prompt engineers! Who would have thought!
I can vividly remember the last few weeks in December 2022 and January 2023. Not a day went by without social media hyping this very role as the job of the future.
Using the latest version of ChatGPT4 as a basis, it is easy to see that the bot can be instructed to generate a prompt for a specific question.
Outsourcing Provider BPO
Business Process Outsourcing (BPO) is a special form of outsourcing and refers to the outsourcing of entire business processes. BPO thus differs from other forms of outsourcing in that it is not a part of the organizational structure, i.e. an organizational unit or department, but a part of the process organization that is outsourced. As a rule, the underlying IT system is outsourced along with the business process. The term Knowledge Process Outsourcing (KPO) is also used.
The outsourced support processes are the actual core business for the BPO service provider. By standardizing its own core processes, the service provider can achieve efficient cost optimization with a low error rate. Optimal utilization of the company's own systems and processes creates economies of scale, which again represent cost advantages for service providers. This creates a win-win situation for the outsourcing company and the service provider.
In the face of the post-modern business reality using ML and AI, however, this will only occur if the process is handled predominantly or entirely with AI tools.
If the service provider can then also use blockchain technology here, for example, to shield accounting as far as possible against incorrect bookings or criminal acts, a real win-win situation will be able to arise.
Finance and accounting, HR/personnel, procurement and logistics looking at Excel in the old way will probably soon be a thing of the past. BPO service providers who force their clients into narrow systems that suit them without mapping the complexity of customer value creation - their days are probably numbered.
In this context, I would not want to miss the opportunity to once again recommend our blog on the topic of industrial asset-light models to the attentive reader.
Link:
Asset-light models - trend or industrial information bubble and echo chamber?
Current observation
There is excitement (..) in the box - that is, between the white-collar professionals.
For the first time in history, a digital technology is emerging that threatens the very repetitive work of these professionals.
What is the answer?
Suddenly, development moratoriums are to be introduced globally. Thetas the wish … . Lobby groups move to impose national bans. They are so quick now … . As much as some of these arguments should be considered in any case, however, the panic cannot be ignored.
Should we go along with it?
Of course, the arguments need to be considered; the AI and ML developers need to respond. Here, too, there probably needs to be some re-sharpening.
But, dear white collar workers, you won't get rid of this!
And that is good!
