You can’t turn a page in a paper or magazine or read a technology blog or email without stumbling upon some commentary on Artificial Intelligence (AI). The mood on the ground swings from the positive perspective of how exciting an opportunity AI is to the negative concerns of people worrying about AI ruling the world or taking their jobs. Are these concerns founded or unjust?
AI covers a broad spectrum of technological solutions but broadly means a system that learns. Consider a ‘smart’ light bulb. If it turns on at 8pm and off at 10pm, that is programmatic, however if it learns your behavioural patterns and learns when you come home and turns the light on based upon what it has learned from your activities, then that is considered AI. This ability to learn can be supervised or unsupervised. Supervised learning is where the system is pre-trained with data and unsupervised learning is where the system is fed or streamed data and then derives its own conclusions. AI can be trained to perform narrow scope tasks very well, much better than a human, but AI systems do not have the breadth of skills a human has. For example, an AI image processing engine can, whilst being fed images of cats over a period of time, very quickly identify a cat in a picture. However, it still does not know what a cat is, certainly not in the way a human does.
John Searle created a famous thought experiment, called the Chinese Room , that demonstrated this fact. His idea was that if one imagined a room with two slots in the wall, through the first slot a piece of card is presented. The person in the room sees it has a Chinese symbol on it. This is followed by a second card with a different symbol. The person pushes them through the second slot and gets a green or red light depending upon whether the sequence is correct. The person learns the right order to get a green light and can repeat the process again and again. Eventually a third symbol appears on a new card and the person in the room tries different combinations of the three cards until they get a green light again. This continues until they are correctly posting a written sentence through the second slot. From this it can be concluded that whilst they know the correct sequence of symbols, they still don’t understand any Chinese dialects. This is a good metaphor for how AI works.
In effect, AI is just a sophisticated programme. It is not sentient or sapient. AI is just a revolution in computational statistics, not a revolution in intelligence. Understanding this helps understand the impact AI will have on peoples’ working lives.
ChatGPT is the heavy lifting text processing software tool and the AI revolution is affecting those that write and publish articles – hence the extreme noise. Yes, AI will be able to replace some human activity in many respects. For example, translating reams of legal documents to locate the correct guidance for a specific case. However, it is NOT replacing the lawyer, just automating a tedious part of their job allowing them to focus on more important things. The problem is compounded by the fact that AI technologies such as ChatGPT seem so human.
AI appearing so human is no coincidence. This is because AI has been programmed by humans. It’s human nature to make things in our own image, it makes us feel more comfortable and secure. This is what makes tools like ChatGPT sound alive. However, AI is far from alive. It does not care or have genuine empathy. AI is devoid of human emotion, which for some tasks can be good but for everyday life is not. AI will not be able to operate effectively in a human world. Just consider how much better dogs and cats have done assimilating into our world with their human-like traits, versus an alien cockroach.
AI is still not self-aware and does not have human cognition. For example, AI does not detect emotion, humour, or irony. It may finally be a technology that is able to pass the traditional Turing test of being able to hold a conversation with a human and the human not knowing if it is a person or a computer, which ChatGPT achieved recently. To that end, new, better tests are starting to be devised. For example, AI cannot pass the Minimal Turing Test.  This test was carried out at MIT (Massachusetts Institute of Technology) where 1,000 participants were asked to choose one word that could only come from a human. This resulted in a list of 428 distinct words. Many obvious words such as banana, love, soul, alive, compassion and empathy were common. When a different participating group were asked which words they thought had been generated by a person or a computer, within these words, whilst getting a high human-ness score, there was one that scored the highest, a score way above all other words. That word was ‘poop’ or ‘poo’. The hypothesis is that this word stands out as being so human because it harks back to our childhood, something that an AI engine never has. This is because humans are social animals, and this is how we learn through shared memories, humour and joint experimentation.
Engineers and researchers are grappling with the challenge of programming AI systems to make split-second ethical decisions when faced with unavoidable accidents or complications. I recently wrote about this specific problem in “Can Autonomous Systems Have Dirty Hands?”, where I concluded they could not, because they do not feel regret. In short you cannot train AI for such scenarios, there are too many factors to consider of which many are not empirical, but social or emotional. Furthermore, in research programmes exploring Phillipa Foot’s famous Trolly Problem,  where people from all around the world are posed the same problems with the differing combinations, researchers have found they get wildly differing answers. This is because ethics are not universal but vary from culture to culture. As humans we naturally negotiate these ethics and what is important to us in social interactions. This is something in life, business and work that an AI engine cannot do.
In summary, yes AI is intelligent, but only because humans are intelligent. It can only learn what humans program it to learn and is restricted by the human’s intelligence. If you are worried about AI ruling the world or taking your job, think of the guy who used to lug boxes around but now does the same job with the help of a forklift truck. AI will not replace us; it will make us more efficient and allow commerce to grow. The problem is not whether AI takes over the world, but will humans force it to take control through exploitation of the technology. As has always been the case, rogue human behaviour is the potential enemy, not AI.