I have a strong feeling that we have a growing mischief in the AI world associated with LLM (Large Language Models). The genie of speculation has been let out of the bottle by OpenAI (or Microsoft) and the subsequent use of neural networks by marketers and students which is always a dangerous combination.
Image 1: Human Brain vs AI Neural Network (Author)
Let's clarify, or rather think together about what is currently happening in AI, and what it may mean for the future in the following areas:
In data terminology, we distinguish Machine Learning (ML), which works on the principle: "Data first, then algorithms", then Expert Systems (knowledge bases made up of human experts) and then AI, which is divided into: Supervised, Unsupervised and Reinforcement learning. With AI, it is very much about what kind of back-propagation (reward, gradient) we use to accelerate the learning process or, if you like, the acquisition of "intelligence" by the machine.
I don't want to rewrite the textbooks, but I don't think this division is enough anymore. Personally, I prefer the original division into deep and shallow intelligence, looking from the outside and from the inside, given the current inadequacy of the Turing test and what the user market "subjectively" thinks about AI intelligence (see The Chinese Room Paradox). However, I see the categorization of AI into Narrow (or Specialized, which now includes ChatGPT), General (AGI) and Super (or Conscious) AI as the most current.
In medicine, the concept of consciousness is described as a state of a certain activation of the brain, which is a basic prerequisite for the emergence of psychic manifestations in humans. Psychologists often speak of consciousness as experiencing (all mental events), awareness of the experienced (re-experiencing the immediately experienced), and awareness of oneself (as experiencing). Broader philosophical and theological perspectives work with multiple concepts: Ghost (or Spirit) that operates outside the body (universal intelligence, creator or "anima mundi"), soul (the embodiment of ghost in the matter of the body), mind (acquired through learning while engaging the senses and interacting with the environment), body (not just the brain), but the very "intelligent" and partially unexplained cooperation of a large number of cells forming more complex structures and functions.
The philosophical concepts of free will, decision-making, and responsibility (awareness of the consequences of choice) are certainly worth mentioning. Also interesting is the psychology's view of the layers of personality and consciousness control (ego, superego, id) and the role of the subconscious, unconscious, attention, and instincts, which are nowadays classic concepts used in medicine, cf. Sigmund Freud & William James. A more comprehensive explanation of conscious and unconscious human action according to C.G. Jung works with concepts such as archetype, shadow, repressed memories and collective consciousness or unconscious. C.G. Jung's concept is quite close to the Eastern philosophies working with the balance of soul and body and the imaginary axis of causality: impulse, thought, emotion, biochemistry which results not only in human health (homeostasis as a state of balance of the body), but also different levels of consciousness (e.g. according to David R. Hawkins). For our purposes of understanding the limits of AI, I would say that there is not yet a multidisciplinary consensuson what all falls under the notion of human consciousness, which is obviously very broad and layered.
In particular, where can human consciousness be found? (Brain, body, out of body), is consciousness acquired? (Learning, growth and interaction) or to what extent is it innate? (e.g. by genes or evolution). However, if we try to compare AI to the limited intelligence of the mind, which is fluid (innate) and crystalline (learned) in nature, then AI can probably catch up and surpass humans on the level of the mind in theory. The learning process of AI (e.g. LLM) is often similar to that of a child only it can be accelerated in some special areas depending on the available training data. As far as the level of complex consciousness of a human is concerned, in my opinion it cannot be compared, if only because humanity has not yet agreed on this concept. However, according to the insight above, the concept of human consciousness is much broader than just "human HW, SW" and we are probably at the beginning of a very interesting debate about what it means to be human, thanks to the confrontation with new AI capabilities.
Image 2: The human brain concept (Author)
It was the application of ChatGPT by students and marketeers for generative tasks like text creation, drawing images, animating videos, creating music...etc. that opened a stormy debate among the public. It's about the new understanding of the concepts of consciousness, creativity, intelligence and whether existing LLMs can ever have abilities and rights as humans. Another chapter being discussed is which tasks can be algorithmized and who AI will replace at work today and tomorrow. I acknowledge my miscalculation. I didn't expect that the first to be replaced would be lawyers, creatives, programmers and if not protected by legislation medical doctors i.e. universally well paid professions. I still think it will be the average and slightly above average people in those professions who will be replaced, but let's be honest with ourselves there are 80% of us in every industry.
My limited insight into neural networks (ANNs, CNNs, RNNs) and the problems with their training datasets, memory, recursion and sequences tells me that there are still and will be many tasks where these techniques will have a fundamental problem getting to AGI or even Super AI.
I am thinking in particular of complex tasks combining multiple subtasks such as language processing combined with mathematical computation, which is more a matter of a multi-step logical procedure based not only on data as a statistical answer. A separate task for me is a unique area of the human body e.g. human sensory such as touch and sensitivity in the hand derived from fine nerves combined with other senses (e.g. sight, hearing, smell). Therefore, I see the path to super (conscious) AI as still unsolved. But I am no longer able to rule it out, and if I see it anywhere it is in the area of "Composite AI" i.e. putting different AI-ML principles together into multiple distributed models by multiple owners (not one Skynet). I would also expect a resurgence in less applied AI-ML practices such as Evolutionary Algorithms, and other NLP (Natural Language Processing) tasks than just LLM.
I can understand the students' enthusiasm for having ChatGPT write a term paper on any topic for them (see LLM principle), but the admiration of the marketeers escapes me. Because here, something that I don't understand in principle and is on average better than I am should be rather feared. However, B2C rightly has a new theme for social, and that's a world I again don't understand.
Honestly, I know I don't know which is the only correct answer since Socrates. I can share my guess on the use of Composite AI i.e. a complex combination of Neural Networks, Evolutionary Algorithms and classical ML (regression, decision trees...etc.) and various other i.e. ethical superstructures (legislatively enforced beyond the basic model, e.g. Responsible, Explainable AI). I noted the Alpaca project i.e. a ChatGPT-like model created by computer scientists at Stanford University costing less than $600 which gives hope for the field of democratizing AI. However, Alpaca, based on Meta's LLaMA system, has so far been withdrawn from the internet shortly after the demo due to concerns about the safety and cost of the project.
Perhaps it is that data oligopolies like Google, Microsoft, Meta will maintain our digital doubles as our personal behavioural model so that they can deliver personalised services to us (the current reality). And we, the customers, will now defend our privacy in the form of a paid personal firewall, where it is not only about privacy, but also about distinguishing which content is man-made and which is machine-made. Thus, in the end it may be a battle of the big but flat tool (ala Transformer LLM, over gpu hw) and our small but personal and precise tool (ala Alpaca, over cpu hw). Oh, and we're probably going to see a lot of regulation too, especially in the EU (see the AI act proposal), we'll see, I'm just predicting without a crystal ball...
Image 3: AI Neural Network (Author)
Provocatively, I would say that in B2C it's already been decided, and in B2B it won't happen for a long time, which is why I started Omnicrane and I only want to do B2B with AI.
On a more serious note, I believe that being human is more than just solving problems flawlessly and that creativity is not just an analogy. And also, that relationships is something that people, and especially business people, will never let take away. The role of a salesperson, in my opinion, is to build a community of customers based on personal relationships and usefulness. Relationship to me is about personal interpersonal affection that needs to be constantly reaffirmed through sharing. And I understand usefulness to mean that I can solve a customer's specific problem better (faster, cheaper, better quality) than they or their competitors. What AI-ML technique I use to solve a given customer's task I consider the only right question in B2B today.
Philosophers haven't yet agreed on what consciousness is and so, it's hard to say when we'll have Super Conscious AI. What is clear, however, is that technological developments in LLM and a number of AI-related fields are moving us forward rapidly in the AGI field, and there will be more and more tasks that AI can solve better than humans.
So, the practical advice for life is whatever you do don't be mediocre at it or you will soon be algorithmizing. And also build communities and relationships based on being real people not avatars or machines that don't make mistakes. Because the combination of human and machine based on algorithms and humanism is, in my opinion, currently the main way to our positive future.
About the author:
Richard A. Novák is an expert in ethical AI, digitalization and IT management. He graduated from the Czech Technical University in Prague and received his Ph.D. in Big Data Ethics from the Faculty of Informatics and Statistics at the University of Economics in Prague. He founded the Prague Data Ethics Lab with his colleagues and teaches courses related to Ethical AI and IT Governance at the VŠE. He founded and currently serves as CEO of Omnicrane, a startup focused on applying AI to SalesTech and MarTech. Previously, he held the position of Vice President at telecommunications companies T-Mobile Czech Republic, GTS Czech and Director of IT Services at Slovak Telekom.
While many B2B traders' sales are stagnant, others are more successful than ever.