Three white nights. This is the time it takes to tame artificial intelligence, promises Ethan Mollick. 49 years old, playful look, discreet style, this management teacher from the prestigious Wharton School (Philadelphia) is the AI teacher we all dreamed of. For more than two years, he says on x and his blog One Useful Thing With clarity and humor the experiences he conducts with Chatgpt and his heirs. A work that propelled him into the ranking of 100 most influential personalities of the venerable AI Time. His book Co-intelligence. Live and work with AI, Which appears in France on March 13 at First Editions, is already a bestseller in the United States.
“Generative artificial intelligence is the exact opposite of what SF’s books and films had described,” he confides with amusement. It was thought that the AI would be precise, cold, gifted in mathematics. Finally, they are Creative, unpredictable and talkative. It is enough to tell an AI that it is wrong to measure the irrationality of the animal. She is able to change her mind by apologizing flatly, even if you tell her stupidity. Or to camp, on the contrary, vehemently in her position when she is wrong.
It is not surprising that so many people hesitate to approach this strange creature. The key to tame it? “Testing AI on an area that we master, advises Ethan Mollick. It allows you to understand very quickly what she can do well and what she fails.” Are you an electrician, lawyer or researcher? Ask him traps. Ask for drafts, quotes, e-mail. The exercise created a feeling of dizziness, generally followed by a great relief. Vertigo at the sight of the tasks that AI performs brilliantly. Sootation when it is seen falling into childish traps or answering banalities.
Get the generative AI
To properly use the generative AI, pleasant Ethan Mollick, “you have to commit a terrible sin: to speak to it as if it were aware”. This expert, including Reid Hoffman, co -founder of LinkedIn and Inflection AI, sings praises knows that consciously, the generative AI is not. These tools are ultimately only tremendously sophisticated probabilistic calculation systems. After having ingested many of our productions (texts, images, etc.), they identify recurrences, patterns. It is for this reason that AI seems so well to understand our questions. And when she claims to be “in love” from us or “wanting to scratch humans from the card”, it shows only one thing: our texts, which she studied, contained similar interactions.
However, it is useful to “pretend”, explains Ethan Mollick. The generative AIs having been drawn to human productions, it is indeed by speaking them in a natural way that they lead them to make the most relevant correlations. “Giving them a clearly defined avatar gives very good results,” he explains. Asking an AI to get into the shoes of a teacher responding to MBA students will produce, for example, not at all the same Results that if he is asked to embody a circus clown. “
Should we speak politely to your AI? It is preferable, confirms the teacher but “without circumlocutions. If you think that the request you make at AI – what is called a “”prompt“” – would be judged nebula by a human, it is likely that AI also understands it badly. “Internet users will also benefit from explaining to AI, step by step, what they want to see them do. And for important or recurrent tasks , to expand their prompt By giving examples but also counterexamples.
The author urges novices to try, from the start, the most powerful AI on the market. “A world separates the latest versions from those last year. Those who test low -end models will miss what AI knows how to do today.” Precepts that help its students imagine ingenious AI uses, such as the creation of “customer people”, these fictitious representations that the marketing services of many companies use to guide the development of their products.
Delegate unnecessary tasks
Where does AI prove the most useful? Mollick now has a fairly clear idea. And encourages it to use it in the fields where the accuracy of AI’s response can be checked quickly. The AI is regularly wrong, this step is indeed key. Chatgpt and its fellows also excel in creative stains. We can ask them 30 ideas for marketing slogans. And repeat the maneuver until a proposition hits the thousand or germinate a good idea. “These news AI also know how to decline content very well in different formats: a company that implements a new internal policy will, thus, quickly produce dozens of training documents adapted to different audiences, with examples, styles of writing and degrees of complexity adapted to each of them. “
Ethan Mollick also encourages Internet users to delegate, at least in part, at AI all these unnecessary tasks that pollute their agendas, like these pompous reports that no one reads. “The AI can also give a useful second opinion. You give it access to data from your problem and you see if it arrives at the same conclusion as you.”
If he studies closely the impact of this technology on the economy, this teacher is mainly at the forefront of the educational revolution she starts. “It is the apocalypse of homework at home,” he observes, “smiles. We are there: artificial intelligence has reached a level which allows it to carry out most of the pupils most Exercises. ” In 2023, already, 82 % of US at first cycle used this technology for school tasks. An upheaval too often interpreted by way, deplores Mollick
“Many teachers wrongly think that this problem can be resolved with artificial content detectors. These tools do not work well. These are the learning and evaluation methods that must be rethought entirely.” Students, for their part, do not always have the feeling or the intention of cheating, observes the expert. But they do not realize that AI maintains a feeling of illusory competence in them. “They get advice, answers that help them do their homework, which give them the feeling of mastering discipline, but removing mental effort, they miss the real objective of education, and Do not imprint in a lasting way, the knowledge they however wish to acquire. ” The key, according to him, is not to use AI to “replace” the reflection, but stimulate it. “By asking the AI to act as a tutor, a coach who guides your steps, rather than a machine to provide responses ‘ready to use’.”
Ethan Mollick has also made teachers from around the world a panoply of prompt IA with an educational vocation. We find there, in bulk, a simulator that teaches us to negotiate, mentors who prick our reflection, a “critical” AI which assesses the quality of a work or even a “pupil” AI to whom a student can train To learn what he thinks he has retained. What inspire the most lethargic students. Word of honor, Ethan Mollick does not let it note their copies.
Co-intelligence. Live and work with AI. Ethan Mollickat First Editions. Publication on March 13.
.