Suggestions
Cassie Kozyrkov
CEO at Data Scientific and Google's first Chief Decision Scientist
Cassie Kozyrkov is a prominent figure in the field of data science and artificial intelligence, currently serving as the Chief Executive Officer at Data Scientific, a stealth AI startup based in New York City, since September 2023.13 Before this role, she was notably recognized as Google's first Chief Decision Scientist, a position she held for nearly a decade, where she played a pivotal role in the development of decision intelligence and trained over 20,000 employees in data and AI applications.13
Professional Background
- Data Scientific (CEO, September 2023 - Present)
- Google (Chief Decision Scientist, 2013 - 2023)
- Federal Reserve Bank of New York (Innovation Advisory Council Member, March 2023 - Present)
- Chief (Chief Member, August 2023 - Present)
- intelmatix (Board of Advisors Member, August 2023 - Present)
Contributions and Influence
Kozyrkov is noted for her influence in decision science and has become a sought-after advisor in AI. She is also an active speaker and writer, sharing insights on decision-making and the implications of AI in various forums.12 Her work has been recognized widely, earning her the title of LinkedIn Top Voice in her field.3
Recent Developments
In October 2023, Kozyrkov announced her departure from Google, describing it as part of a personal commitment to not remain at any company for more than ten years. This transition marks a significant shift in her career as she embarks on new projects aimed at leveraging her expertise in decision intelligence.34
Kozyrkov's career trajectory reflects her dedication to advancing the understanding and application of data-driven decision-making processes across various industries.
Highlights
Why might a leader in charge of building an AI system care about how we navigate today’s complexities regarding diversity?
Because there's not one right way to build a complex AI system. Not ever.
The right way depends very much on what the AI system is for... and WHO it is for.
If you care about developing technology in an ethical and responsible manner, it’s your duty to think about all the people your solution could affect. Thinking about them after you’ve built it is irresponsible. The time for it is when you start! And thinking about them isn't enough -- ensure that the people affected by what you're building are given a voice.
Too many people forget that it's the leader's job to make some very subjective decisions that shape the nature of any AI system. Decision-makers must make judgment calls for which there is no right answer, but often there are many wrong answers... especially when the impact lands disproportionately with a minority of the population.
All technology always comes from and is designed by people, which means AI is no more objective than we are. AI is a big fat subjectivity sandwich with a thin slice of objectivity.
A layer of math and data spread on top of the host of utterly subjective choices (What shall we apply AI to? Is it worth doing? In which circumstances? How shall we define success? How well does it need to work? etc.) obscures the ever-present human element and gives an illusion of objectivity.
It’s much flashier to say, “The AI learned to do this task all by itself…” than to tell the truth: “People used a tool with a cool name to help them write code. They fed in examples they considered appropriate, found some patterns in them, and turned those patterns into instructions. Then they checked whether they liked what those instructions did for them.”
The truth drips with human subjectivity — look at all those little choices along the way which are left up to people running the project. Wrapping them in a glamorous coat of math doesn’t make the core any less squishy. Whenever math is applied towards a purpose, that purpose is still shaped by the sensibilities of our times.
That's why so much depends on how the decision-maker sets the priorities. There's a lot they need to consider, especially when it comes to the complexity of an AI system, and it's an awful burden to try to represent everyone all by yourself. That's how a leader fails at this.
The way to succeed is to ensure that you're inviting a diverse set of perspectives to help you see the things you might miss otherwise.
Think of data as textbooks for the AI system. Like regular textbooks, dataset have human authors. A good teacher knows these textbooks need to be checked before they're shown to the student, no matter what it says on the cover. A great teacher knows that one person's review is not enough when the textbook deals with anything important. It's important that datasets, like textbooks, are checked from diverse points of view.
OpenAI's new model o1-preview (aka Strawberry 🍓 ) is here! If you'd prefer not to use up your 30 preview credits, I've recorded a video walkthrough for you comparing my attempt to bait the old model and the new model into the same hallucination: https://lnkd.in/eB4xEasi
Hallucination, reasoning, thinking, and knowing are just some of the words we should be careful about when it comes to our own reasoning about AI systems. There's no thinking happening here, just processing. Processing what? Consistency checks. Instead of simply generating output, Strawberry is built to review what it's outputting (multiple times in parallel) for consensus and consistency.
If you found the video useful, drop me a note in the comments so I know what to do more of next time! ❤️
