Artificial intelligence: myths and reality
Artificial intelligence (AI) is a buzzword that has been on everyone’s lips in recent times. On the one hand, it seems to be an almost magical way of solving all sorts of practical problems. On the other, some are skeptical about its effectiveness—while others fear it may lead to the outright self-destruction of humanity.
What is the reality? The methods of artificial intelligence, or to be more precise, machine learning, and especially its latest form, deep learning, are a real breakthrough in computer science. We can see this with our own eyes: computers and mobile devices recognize trees, elephants, tanks and ourselves in pictures, understand what is being said to them, and translate texts no worse than we would ourselves.
The applications of artificial intelligence do not end with individual consumption. They are already transforming the world of business and administration. There is no industry that will not be irreversibly changed by AI solutions. When using these solutions, however, it is easy to make all sorts of mistakes or fall into deep traps.
CSI Expertise, or how to implement artificial intelligence
The Artificial Intelligence Center at AMU is helping companies and institutions to introduce AI solutions into business practice by offering a unique service called “CSI Expertise.” Its purpose is not to offer specific artificial intelligence systems, but to help choose the right AI solution. This is preceded by an analysis of the organization’s maturity and readiness to implement artificial intelligence solutions.
Imagine an executive in a company or institution. She analyzes a business process, and comes to the conclusion that its automation will dramatically reduce costs or open up entirely new markets. Traditional IT methods, however, are not enough. Our manager wonders if perhaps artificial intelligence, which is so prominent in the media, could help in the transformation process. She organizes a bidding competition for AI companies, and a mixed bag of bids flows in. What selection criteria should she adopt? Isn’t there a risk of being fooled by the fine words and incomprehensible jargon of the bidders? How to make the 99% efficiency results presented in the bidders’ tables and charts translate into real success for the organization?
This is where the AMU Artificial Intelligence Center can help. Its experts will take care of the proper preparation of the bidding competition, and advise on what is possible and what is beyond the scope of current technology. They will also help in the process of testing solutions and selecting the best one.
Artificial intelligence: useful or harmful?
A key issue is the evaluation of artificial intelligence solutions. Wrongly constructed IT solutions can lead to absurd and harmful decisions. In the case of artificial intelligence systems, the risk of harming the user is particularly high. This is because machine learning systems rely on “brute-force” optimization of a given objective function. If the objective function does not align with the economic interests of the organization, the result can be disastrous.
The problem of misspecification of the objective function can be illustrated by the parable of the “paperclip maximizer.” If, in a multi-industry factory, we instruct artificial intelligence to organize the production process in such a way as to maximize the number of paperclips produced, then, in the event of “success” of this undertaking, we may wake up in a world where... every radiator and every car has been turned into paperclips.
Gonito – or how to objectively evaluate artificial intelligence
Experts at the Artificial Intelligence Center will help organizations choose the right way to evaluate proposed solutions. CSI is developing Gonito, an open-source evaluation system that enables the accurate evaluation of AI systems under controlled conditions, so as to avoid data leaks and the overestimation of results—whether conscious or unconscious—by bidders or contractors offering AI solutions.
How to avoid the pitfalls of artificial intelligence?
The assistance of CSI specialists does not end with evaluation. As part of the AI Expertise service, we educate and advise organizations on how to avoid a variety of pitfalls associated with implementing AI-based processes. Problem areas include: the resilience of systems to time drift (did the system stop recognizing dates at the beginning of 2022? can the system cope with new tax laws?), data confidentiality (can one user’s data leak unnoticed to another through the AI system?), security (can the AI system be abused by a determined attacker?), and interpretability (why did the AI system give this answer and not another?).
Artificial intelligence without borders
We are the conduits of a new world in which artificial intelligence will touch every sphere of life. Searching through archives? Counting blueberries from photos? Detecting burglaries? Predicting bicycle sales? Classifying correspondence? Supporting language teaching? It’s hard to find a field where there would not be a place for AI solutions.
Today, every entrepreneur must find their way in this new reality—full of opportunities, but also risks. We help minimize those risks.