Yanay Zaguri - AppsFlyer

"The AI genie is out of the bottle and I don't really think it can be put back in"

Yanay Zaguri, an expert in learning technologies and AVP of Organizational Growth at AppsFlyer, tackles the artificial intelligence revolution: "Humanity needs to write a new code of ethics"

In the workshops he conducts, Yanay Zaguri likes to show the students a slide with 15 of his photos. One authentic photo, taken by a real-life photographer, and the rest which were all created synthetically using Stable Diffusion, one of the popular AI-based image generators. He asks the students to determine which of the pictures is the authentic one. They almost always fail.
Zaguri is a learning technology expert who has been involved in learning in organizations for 25 years. He wears two hats, he is a faculty member in the Faculty of Learning Technologies at the HIT Institute in Holon, and he is also AVP of Organizational Growth at AppsFlyer, which deals with the measurement and analysis of mobile marketing campaigns. Zaguri first approached the world of generative artificial intelligence as a private user, eager to play with a new toy. Then, he started thinking about how this technology can be harnessed for the benefit of learning.
1 View gallery
ינאי זגורי מנהל טכנולוגיות למידה ב HIT
ינאי זגורי מנהל טכנולוגיות למידה ב HIT
Yanay Zaguri - AppsFlyer
(Photo: Orel Cohen)
"Organizational learning experts are interested in one question: how do we change behavior," he says. "A culture of learning in organizations is not for the sake of education. We want them to sell more, to be more helpful, to give more value to customers, for the managers to be better to the employees. And in this place technology fits in perfectly, because it can help us shorten the learning process and make it a lot more effective."
Can you give me an example?
"Let's say I have a salesperson who needs to deliver a message to the customer that they don't necessarily want to hear. A message that contains a refusal. With the help of AI, you can draft an email to the customer that will be on the one hand decisive, and on the other hand inclusive. They will tell the customer in the best way that they cannot accept the 25% discount that they requested, and at the same time they will express a desire for future cooperation, and make it clear to the client how important they are."
It sounds like the AI doesn't teach the salesperson, but replaces them.
"When we talk about professions that will be affected by AI, we need to distinguish between two areas. The first is a language model - text to text. ChatGPT and others are in this field. You feed it text, and it returns text to you. This model understands language well, so it demonstrates very high rhetorical skills. This does not mean it has a command of the facts - it can present misinformation very convincingly.
"The second model we see is text to image. It's a model that knows how to make an excellent connection between language and images, and to create a new image based on lots of other images."
So where does that leave our salesperson?
"Anyone who uses language directly in the context of work will be affected. Whether it's salespeople, whether it's advertising and marketing people, whether it's journalists, whether it's content creators - everyone who uses language as a central tool within their work method. And when I say language, I mean any language. Code is also language. Music too."
And on the other side, everyone who deals with media?
"Yes. The second group is everyone who deals with rich media. Photos, movies, sound, graphics, video editors, after effects. All of these will undergo significant changes."
Will these professions disappear?
"I don't believe that professions will disappear completely. And even if they do, it will take a long time. Another thing that will happen is the transition to a supervisory position. The machine will do the work, and people will monitor that it is not doing nonsense, or will take action in cases where the machine encounters a problem that it does not know how to solve.
"But there is another important thing. How do you acquire a profession today? You go through a repetitive learning process. Let's say you have an intern at a law firm. The intern draws up one contract and then another, and drafts again and again. In the world of AI, the veteran lawyer can request everything from the machine, instead of from the intern. This means that one major change we are expected to see in the world of employment will be that junior positions will disappear. Here we really start talking about new economic models. Because it's not just what the junior's capacity is and what they are capable of doing, it's also how much you pay them. And, it raises the question - how do you become senior, and what does it mean to be in a senior position."
The rapid rise of generative artificial intelligence raises many concerns, from privacy issues to fake news. "ChatGPT allows anyone to write quality content. They can create content that in advance aims to influence a certain opinion or speak to a very specific sector. The level of targeting increases wonders," says Zaguri.
It sounds like a paradise for fake news creators.
"This technology comes with a threat, like any technology. If you look back at history, every technology had a wave of people who misused it, then a counter wave of barriers was created, and a game of cat and mouse began. This will also require regulatory intervention, and the tech companies will also have to find answers."
Tech companies are not known for protecting users.
"They will have no choice. In social networks, for example, they will have to train their algorithms to promote certain posts that reflect values of unity and liberality, and hide others that create division. Not because of their kindness, but because otherwise users will leave them en masse, and when there are no users, there is no money. In the end, it's a combination. On the one hand, regulation, and on the other hand, the responsibility of the tech companies."
At the end of March, the world of technology was shaken by an unusual letter sent by about a thousand researchers and experts in the field of artificial intelligence and computer science. In their letter, the signatories warn of the possible dangers inherent in advanced AI systems, and call for a six-month halt to development in the field. Zaguri does not believe that the letter is expected to affect development in the field. "The genie is out of the bottle. It's already out, and I don't really think it can be put back in. I think it's something that is between a struggle for control, and an attempt to call on humanity to understand the magnitude of this period. It’s time to call on the regulator to wake up. To call the 'good' researchers to wake up."
And will it work?
"I’m not sure. It may be that this attempt, which is a bit clumsy, is too little too late. But it can lead to thinking about the ethical questions behind the scenes. Humanity needs to write a new code of ethics, redefine where the red lines are drawn."

The art of writing the prompt
A prompt is an instruction given to an artificial intelligence tool, as a starting point for creating new content. A prompt can be a sentence like "Write a story for children about a magical adventure", or "Write me code for a recipe website". The more accurate the prompt, the better the results. But how do you write an accurate prompt? Zaguri enumerates a number of guiding principles:
1. Progress in stages: As long as we correspond in the same conversation window, the chat "remembers" the context. Therefore, one of the simplest things is to start from a general text description, and as the conversation progresses to improve it and become more specific.
2. Tone and voice: Clarify the prompt with tone and voice. An approach regarding garbage removal in a building, for example, can be requested once in an official tone, if applying to the municipality, and once in an informal, lighter tone, if applying to the neighbors. Another thing that the chat understands very well is to write in a certain voice. For example, to formulate a certain text as if Joey from "Friends" said it, or President Barack Obama.
3. Specification: Don't just tell the chat to give you a plan for a trip to Greece, but ask for a plan for a trip to Greece that is suitable for a family with two children, that plans to rent a car in a certain area, which will also include a relaxing day on the beach. On top of that you can add restrictions - what you don't want the trip to include.