Elad Ziklik recently received a parking ticket in his hometown of Kirkland, the sleepy suburb of Seattle where he has lived since moving to the US. The last time this happened, he hired a lawyer who drafted a letter of appeal to the municipality for him. This time, Ziklik turned to ChatGPT, which within minutes produced a letter which ultimately acquitted him of the pesky fine.
Ziklik is no stranger to artificial intelligence. As Oracle VP AI and Data Science Services, he is the most senior person in AI at the company which is slated to shape the way generative AI will be made accessible to hospitals, police and security organizations, banks, and government offices.
"I saved a hundred dollars," Ziklik tells Globes in a Zoom call from his home. "And that’s a problem for lawyers, because there are probably many more like me. Any lawyer who doesn't sit down right now and think about how their business will change as a result of generative AI, will be left behind," he warns. But despite Ziklik 's respect for OpenAI, the AI company behind chatbot ChatGPT, when it comes to the big money from the business sector, Ziklik is still skeptical. While many people have gotten used to using ChatGPT for personal needs such as transcribing articles, or making plans for a family vacation, their workplaces still shy away from it. For Ziklik, who first encountered OpenAI three years ago, while working at Microsoft, this concern is understandable.
"The black box offered by OpenAI today presents a serious problem," he says. "This reality - where a smart model offers me something and with that, my ability to play with it ends, and I have to trust it blindly - may be relevant for writing love songs for my partner or other cool uses - because in those cases, you don't really care where the products come from. But as a company or a business, I can't afford to use a product that someone else has a copyright on, or that it does something wrong and I’m not able to fix it, or the person who developed it decided that telling a joke about Donald Trump is not okay but Joe Biden is okay. Because those safety mechanisms - what we call in the professional jargon "guardrails" - were defined by OpenAI’s people without transparency, and without me being able to adapt them to my organization. And that’s the heart of the problem for the solutions they offer.'
It’s not just talk. Last week, Oracle founder Larry Ellison unveiled the project Ziklik has been leading for the past few months: a $270 million strategic investment on behalf of Oracle in Cohere. Known by the US tech media as OpenAI's most significant competitor, Canadian company Cohere was founded by former Google employees with the stated goal of "Google-standard AI."
"In some aspects, Cohere has several advantages over OpenAI," Ziklik points out. "For example, it will be more difficult for OpenAI to approach governments, convince them to place their confidential data with it, or on Microsoft's public cloud, and build things together that way. Instead of this type of closed model, we can bring Cohere's AI models to government servers, and tell them, ‘We won’t touch your data, because it’s yours, but you will be able to use our heavy models on a scale that suits you, without your having to build a server farm with a thousand graphic processors.’"
When Ziklik outlines business use cases for AI, he usually sticks to examples from the healthcare sector, and for good reason; last year, Oracle completed the largest acquisition in its history: $28 billion for US healthcare computer services company Cerner. "By law, we’re not allowed to offload Cerner's data from our servers, so public cloud models [like that offered by OpenAI] don’t suit us," he says.
Oracle is already conducting pilots at US hospitals that allow doctors and nurses to save hours of work on hospital discharge summaries. "Sometimes you wait two hours in the emergency room for a letter like this. Imagine if a doctor could press a button that summarizes your entire hospitalization, explains your medication protocol to you, and leaves the doctor free to handle other patients, and do what they need to do. As an Israeli living in the US, I’m always surprised to see how busy doctors are with papers, summaries, printing, regulation, approvals with the health insurance companies - the amount of time that’s wasted without treating the patient is simply ridiculous. There’s an opportunity here for a total revolution in the way the health system is run. It's amazing how thirsty this sector is for technology, yet how difficult it is to introduce new technologies because this sector is very conservative. But generative AI has just opened everyone's minds in an unbelievable way."
Isn't it a bit dangerous to have AI generate recommendations for post-hospitalization treatment?
"As someone who’s been dealing with AI for 15 years, I always thought that the more accurate, the better, right? But what if the model is right only 50% of the time? In that case, doctors will know that it is accurate one out of two times, so they’ll check each recommendation seven times. But if the model is accurate in 99% of cases, doctors will take it for granted, and one out of every hundred patients will die because of recommending the wrong medicine or ignoring an allergy that should have been taken into account.
"Therefore, when you create an AI model, especially for a critical field like health, you must create the accompanying mechanisms, controls and guardrails that will enable others to understand the considerations that led you to make a certain decision, and also allow them to critique you.
"This is AI's problem and opportunity: OpenAI will launch ChatGPT 4, 5 or even 9, and it will pass the bar examination but if it does not relate to the commercial laws relevant to the health sector, legal sector, or any other business, it simply will not work. Even today, judges in the US are demanding full disclosure on whether this or that statement was written using AI."
They made him offer he couldn’t refuse
Ziklik (46) served in IDF Intelligence Unit 8200, worked at a startup that collapsed during the dot-com bubble, and was then recruited as the first engineer to work at Microsoft's new R&D center in Herzliya. After five years working on telecom technologies, Ziklik took it upon himself to found a sort of technological incubator in what was then called big data, analytics, and machine learning; smart data management technologies that were then viewed as the most advanced forms of AI. "At that time, Microsoft worked mainly with IT system managers, and we wanted to start talking to other senior VPs within the enterprise, like human resources, marketing, and finance, in order to bring them over to the Microsoft cloud."
The first AI technology application actually came via Microsoft’s game console Xbox; Ziklik proved that it was possible to apply a recommendation engine developed by his group -- which up to that point had only recommended movies and games - to consumer and business solutions as well. The project made him famous throughout the company. CVP and CEO of AI Joseph Sirosh , who was responsible for all Microsoft machine learning, recruited Ziklik and his team to establish what is now known as Cognitive Services, a menu of AI code allowing Microsoft software developers to easily add features into their applications.
After that success, Ziklik decided to accept the offer to move to Seattle with his wife and three children. Then, after 14 years at Microsoft, Ziklik received an offer he could not refuse to manage the entire AI division at Oracle: "I was thrilled to take all the hard-won experience in this field, all the things we - me and others - had done at Microsoft, and bring them to a cloud company that was rebuilding the sector. But this time, I would be at the helm."
A year ago, even before competitor OpenAI launched its image generator DALL-E and generative language model ChatGPT, Ziklik realized that something was happening. "I didn't imagine that ChatGPT would take over the world so quickly, but it was clear to me that this technology would create a very serious disruption to the industry. I’d met OpenAI already three or four years ago and understood where they were going. At Oracle, we met with a lot of companies in the sector; some with teams of geniuses but who don’t really care about offering products or aren’t really built to work in partnership. Our first meeting with Cohere was already in August of last year. We clicked even at that time: we had a common worldview about how to approach different business audiences. Feelers were sent out, here and there, and over the last three or four months we entered into accelerated negotiations until, over the last month and a half, we had daily conversations, including nights and weekends. Unfortunately, or fortunately, I was in on all of them. Larry and Safra Catz [ Oracle co-CEOs] were involved in the details, and I personally went over each of the three contracts signed in this deal. After that, we went to Toronto to celebrate, and to understand how we’ll work together from here on.'
But truthfully, you didn’t acquire Cohere and you didn’t sign an exclusive agreement with them. Your competition, like Salesforce and Google, sit at the same table with you. They can make a similar deal with Cohere tomorrow morning. You don’t have exclusivity.
"True, and I think it's great that many people will use them. There will be things that they will provide to others and there will be things that we will work on in exclusive partnerships, for example, health-based models based on Cerner data, or internal security models based on our work with governments."
Are we in an AI financial bubble?
"There is a bubble, absolutely. That is, there are companies that don't have any customers, don't make a dollar and are raising $300 million at a $2 billion valuation. But I really do think that some will be like the companies who brought us the Internet. There’s certainly a lot of hype and fantastical investments, but is the hype justified? Yes, because the company that manages to get in and take over is the next Google - just as Google took over the Internet with search, they will take over productivity with AI."
We saw that Cerner laid off thousands of people - it looks like people were laid off to make way for bots.
"There are few cases in the world where jobs disappear - they simply change. A Prompt Engineer, a profession that didn't exist before, is a person who learns and knows how to give correct instructions to ChatGPT. The bot generates code, but you have to know how to manipulate the beast to make it do what it has to do."
Published by Globes, Israel business news - en.globes.co.il - on June 27, 2023.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2023.