For many years we have been used to having rule-based algorithms around us, systems that behave exactly as they should and that respond correctly to our needs.
Consider, for example, how a bank transfer works: we enter the requested data and we expect the money to be moved from one account to another, all because the systems follow precise rules with which they have been programmed. It may happen that there is some malfunction, but these are problems that are identified and resolved, so that the system returns to function as it was conceived.
Over time we have also become accustomed to artificial intelligence algorithms, systems capable of making choices independently based on the context and what the machine is able to learn thanks to the use it makes of it. The next song on Spotify, the products recommended on Amazon, advertising on social networks are not at all random, on the contrary they are linked to our tastes, our preferences, our propensity to buy. Nothing is random, everything is guided by algorithms which, day after day, learn something about us and which are designed to provide us with something that can somehow meet our taste or can stimulate the desire to buy something in us.
History of OpenAI, the company founded by Musk and Altman behind the ChatGPT phenomenon
by Archangel Rociola
More recently, with the public availability of tools such as ChatGPT, even non-experts have been able to experiment with the generation of complex texts starting from a huge knowledge base, obtaining a sometimes good quality product made thanks to an excellent of sentence construction. This has made it possible to have content that has tended to be written very well, the quality of which, albeit with limits, appears good, but whose truthfulness is a mystery. This last point depends on the knowledge base which, unfortunately, has favored quantity over quality, thus incorporating a part of correct information and another part of completely wrong information.
It follows that the final result can only be imprecise and, in general, the contents produced cannot be trusted, even if they are written in a very convincing way.
Microsoft expands access to ChatGPT, expected new services for users
by Archangel Rociola
Up to now we have seen algorithms capable of making decisions based solely on pre-set rules in the programming phase, up to algorithms capable of choosing based on concepts and knowledge acquired massively and then at each subsequent interaction.
In any case, the goal of the algorithm has always been to solve the problems assigned by making decisions within a precise frame of reference, without having the freedom to address the issue in a completely different way or to subvert the operating rules. What would happen instead if we had algorithms able to use lateral thinking or to act completely outside the pre-established schemes to perform the tasks assigned to them in the best possible way?
Lateral thinking is a human thinking technique that consists of analyzing a problem from different perspectives, generating ideas outside the box and finding unusual and innovative solutions. It is therefore something very far from the operating logic of rule-based algorithms, in which case there are no degrees of freedom, but only the clear and indisputable sequence of what must happen according to the surrounding conditions.
Using Artificial Intelligence in the relationship with customers
Algorithms based on artificial intelligence tend to rely on a large amount of data present in the knowledge base, but the goal for the neural network is to find a match with one of the solution schemes that the machine knows as valid, so very often the work of these algorithms is limited to identifying reasonable solutions and within sets of valid resolutions. Also in this case therefore, despite the presence of neural networks and very large amounts of data, very often what is obtained are rather traditional solutions.
If we really wanted to exploit the enormous potential of algorithms based on artificial intelligence, we should really move within the context of lateral thinking, trying to free some constraints and allow the machine to propose completely unthinkable solutions. This puts us in front of the possibility that the solutions are technically inapplicable, but it is the risk that it is necessary to run when trying to raise the level of innovation of a system.
On the one hand we could have truly innovative solutions in many fields of application, from finance, to healthcare, to production and sales logics, to marketing, but on the other we would have the risk of obtaining completely inapplicable solutions because they are illegal, because they are not compatible with some pre-existing constraints or, simply, because they are inappropriate or inconsistent with our ethical and moral principles.
Think, for example, of a machine which, in order to win any game, decides to cheat, even if it runs the risk of being discovered. Over time he could learn to cheat better, to the point of refining this characteristic of his and transforming it into a skill at his disposal. On the other hand, if the goal is to win the game, the rules may become less important. In the same way we could have financial algorithms capable of moving wealth with perhaps more efficient, but illegal logics.
Or we could have machines, which we have asked to help human beings better preserve planet Earth by providing solutions to reduce carbon dioxide emissions and pollution, which may consider the extinction of mankind a very good idea which, we know very well, it is the main cause of the problem.
The machine would have “technically” found an excellent solution, it remains to be seen whether mankind would be happy with a solution of this type.