OpenAI CEO Sam Altman on the Future of AI

Summary

In a thought-provoking conversation, Sam Altman, the CEO of OpenAI, discusses the future of artificial intelligence (AI) and its potential impact on society. He shares his experiences from traveling the world, engaging with various stakeholders, and observes a global excitement and optimism towards AI, despite the并存的焦虑 (coexisting anxiety). Altman emphasizes the importance of thoughtful development, global cooperation, and the need for regulation to ensure AI's benefits are realized without catastrophic risks. He addresses concerns about bias in AI, the role of startups, and the economic implications of AI development. Altman also touches on the relationship between OpenAI and Microsoft, the influence of Elon Musk, and the challenges of global regulation. He reflects on his personal motivations, the potential for AI to augment human capabilities, and the necessity of creating a governance structure that involves all of humanity to manage the power of advanced AI systems.

Takeaways

• 🌟 **Global AI Enthusiasm**: There's a high level of excitement and optimism about AI's future potential worldwide, with a more intense interest outside the Bay Area.
• 🤔 **Balancing Optimism and Anxiety**: While there's excitement about AI, there's also a healthy level of anxiety and thoughtfulness regarding its development and societal impact.
• 📝 **Feedback-Driven Development**: The development of AI is being influenced by extensive feedback from users, developers, and world leaders, leading to specific changes in approach.
• 🌐 **Global Cooperation**: There's a call for global cooperation in AI development, with an emphasis on getting AGI (Artificial General Intelligence) right to avoid potential downsides.
• 🚨 **Risk Mitigation**: The dangers of AI are acknowledged, and mitigating the risk of extinction from AI is considered a global priority alongside other major risks like pandemics and nuclear war.
• 📈 **Exponential Growth**: AI is on a steep exponential growth curve, and understanding this growth is crucial for anticipating future developments and challenges.
• 🏆 **Positive Impact Potential**: AI has the potential to deliver significant benefits, such as improving education and healthcare globally and contributing to scientific progress.
• 💰 **Economic Incentives**: Despite the potential for economic benefits, the focus is on ensuring that AI's benefits are accessible to everyone, not just those who can afford it.
• 🏛️ **Regulation and Oversight**: There's a push for global and coordinated regulation of AI, especially for systems that pose existential risks, to ensure safety and prevent misuse.
• 🔍 **Bias and Ethics**: Addressing biases in AI is a significant concern, with ongoing efforts to reduce bias and align AI models with human values and ethics.
• 🌱 **Long-Term Contribution**: The motivation behind AI development is not just financial but also a desire to contribute to human progress and leave a lasting, positive impact.

Q & A

What is the main topic of discussion between the guest and Emily?

The main topic of discussion is the future of AI, its development, potential risks, and the importance of global cooperation and regulation in ensuring its safe and beneficial use.

What has been the guest's experience traveling to various places like Rio and Tokyo?

The guest found the experience very special, as it allowed him to talk to users, developers, and leaders interested in AI. He was surprised by the level of excitement and optimism about the future of AI, which was more intense outside the Bay Area.

What are the guest's thoughts on the current state of AI development?

The guest believes that there is a mix of excitement and anxiety regarding AI. He appreciates the thoughtfulness and nuanced understanding of AI's potential that he has observed globally and the desire to use technology for social good.

How does the guest plan to incorporate feedback from developers into AI development?

The guest has taken extensive handwritten notes and plans to make specific changes based on feedback, including customization of tools to reflect users' values, culture, and history, as well as ensuring representation in AI development.

What is the guest's stance on global regulation of AI?

The guest supports global regulation and believes it is crucial for ensuring safety. He emphasizes the importance of a coordinated global response and suggests that overregulation should be avoided, especially for small startups.

How does the guest respond to critics who say he is calling for regulation in public but lobbying for something else in private?

The guest denies such allegations, stating that they are pushing for regulation privately as well. He believes in the importance of regulation, especially for systems with high risk levels, and dismisses the critics' claims as intellectually dishonest.

What are the guest's views on the potential benefits of AI?

The guest sees tremendous upsides to AI, including the opportunity for everyone on Earth to have a better quality education, improved medical care, scientific progress, and the potential to end poverty. He believes that managing through the risks is necessary to achieve these benefits.

How does the guest address concerns about AI's potential to amplify our worst impulses?

The guest acknowledges that AI can do bad things as well as good, and he does not have an overly optimistic view. He believes that AI can help reduce bias in the world and align models to behave in certain ways, but the question of user control and value system limits will be a tough one for society to address.

What is the guest's response to concerns about the potential misuse of AI?

The guest believes that while there are risks of misuse, the technology has advanced to a point where it can align models to behave in certain ways, reducing the potential for misuse. He also suggests that the limits of user control and the value system are questions society will need to grapple with.

How does the guest explain his lack of personal financial incentive in OpenAI?

The guest clarifies that he has enough money from other investments and does not need to profit personally from OpenAI. He is more interested in making a significant impact on human technological progress and finds great personal fulfillment in the work.

What is the guest's vision for the governance of AI in the future?

The guest envisions a future where the governance of AI is democratized and not controlled by one company or person. He suggests that the board of OpenAI should eventually represent all of humanity and that the technology's benefits and access should belong to everyone.

Outlines

00:00

🌟 Global Perspectives on AI's Future

The guest discusses his travels to various places like Rio and Tokyo, engaging with users, developers, and leaders interested in AI. He expresses surprise at the level of excitement and optimism about AI's future, even surpassing that in the Bay Area. The conversation also touches on the anxiety that comes with the potential risks of AI, emphasizing the importance of thoughtfulness and nuance. The guest reflects on how his approach to AI development might change based on the feedback he has gathered, highlighting the need for customization and representation of various cultures and values in AI tools. He also talks about the importance of global cooperation and the signing of a statement warning about the existential risks of AI.

05:02

🚀 The Transformative Potential of AI

The dialogue focuses on the transformative potential of AI in areas such as education and healthcare, suggesting that AI could lead to globally accessible, high-quality services in these sectors. The guest speaks positively about scientific and technological progress and the potential to end poverty. However, they also acknowledge the need to manage risks. The discussion addresses the economic benefits of AI and the challenges of global regulation, with the guest advocating for a balanced approach that doesn't stifle innovation. Criticisms about the potential for regulatory capture and the call for a certification system for AI are also explored.

10:07

🤖 AI's Impact on Human Behavior

The conversation delves into AI's potential to both enhance and amplify our worst impulses. The guest maintains a realistic view, acknowledging that while technology can cause harm, it has also historically brought about significant good. He discusses the progress made in reducing bias in AI models and the challenges of aligning AI behavior with human values. The guest also addresses concerns about the data that AI is built upon, which can be biased or flawed, and the ongoing efforts to safeguard against these issues.

15:10

💰 Financial Incentives and Personal Motivations

The guest clarifies his financial incentives, or lack thereof, in being part of OpenAI, a nonprofit organization. He explains that he has no equity in the company and that any personal financial benefit from OpenAI's success is immaterial to him. Instead, he is driven by the desire for an interesting life impact and the opportunity to contribute to technological progress. The guest emphasizes that he already has enough money and that his primary goal is to make a meaningful contribution to humanity through AI.

20:13

🌍 International AI Development and Trust

The discussion touches on the state of AI development in countries like China and Russia, with the guest expressing a desire for more information and a collaborative approach to AI on a global scale. He stresses the importance of not relying on a single person or company for such a powerful technology and suggests that the governance of AI should be democratized over time. The guest also addresses the public's trust, stating that trust should not be placed in any one individual or entity when it comes to AI, reflecting the significant responsibility and power that comes with shaping AI's future.

Keywords

💡AI

Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. In the video, AI is the central theme, with discussions about its future, potential risks, and the excitement surrounding its development. An example from the script is the mention of 'AI safety' and the 'development of AI', indicating the importance of responsible AI growth.

💡Exponential growth

Exponential growth describes a process that increases at a rate proportional to its current value, leading to rapid expansion. The video discusses the exponential curve of AI technology, emphasizing the rapid pace of advancement and the challenges in understanding and managing such growth. The script mentions 'We are on an exponential curve and a relatively steep one', highlighting the speed at which AI is progressing.

💡Global regulation

Global regulation refers to the establishment of rules and standards across different countries to govern a particular domain. In the context of the video, global regulation of AI is discussed as a way to ensure safety and ethical use on a worldwide scale. The script includes a discussion about meeting with 'world leaders' and the 'desire for the world to cooperate', which ties into the concept of global governance of AI.

💡Bias in AI

Bias in AI pertains to the unfairness or prejudice that can be inadvertently built into AI systems, often reflecting the biases present in the training data. The video addresses the issue of bias, noting the efforts to reduce it through techniques like reinforcement learning with human feedback. The script states, 'THERE WAS A RECENT STUDY THAT GPT4, THE MODEL THAT IS, RELEASED, IS LESS BIASED ON IMPLICIT BIAS TO TEST EVEN FOR, HUMANS WHO THINK THEY HAVE REALLY TRAINED THEMSELVES TO BE, GOOD AT THIS', showcasing an awareness of the issue and the progress made.

💡AI safety

AI safety involves the practices and research aimed at preventing AI systems from causing harm or unintended consequences. The video emphasizes the importance of AI safety, with discussions about potential risks and the need for precautionary measures. An example from the script is 'A REIT'S OF MITIGATING THE RISK OF EXTINCTION FROM AI SHOULD BE, A GLOBAL PRIORITY ALONGSIDE OTHER SUBTLE SKILL RISKS THAT, JUST PANDEMICS AND NUCLEAR WAR.'

💡Education and AI

The use of AI in education refers to leveraging AI technologies to enhance teaching, learning, and administrative processes. In the video, the potential for AI to provide a better quality of education globally is mentioned as one of the significant upsides of AI development. The script includes a statement about 'the opportunity for everyone on Earth to have a better quality education', emphasizing the transformative potential of AI in this sector.

💡Medical care and AI

The integration of AI in medical care involves using AI to improve patient outcomes, streamline healthcare operations, and enhance medical research. The video discusses the transformative impact AI could have on medical care, making it accessible globally. The script states 'MEDICAL CARE AND WHAT I THINK IS GOING TO HAPPEN THERE, MAKING THAT AVAILABLE TRULY GLOBALLY', indicating the belief in AI's potential to revolutionize healthcare.

💡Elon Musk

Elon Musk is an entrepreneur and CEO known for his work in electric vehicles, space exploration, and tunnel construction. In the video, Musk is mentioned in the context of his concerns and public disagreements with the direction of AI development, particularly with OpenAI. The script references 'ELON MUSK WAS SCARED OF GOOGLE' and discusses the public and private interactions between the speaker and Musk.

💡OpenAI

OpenAI is a research laboratory that develops open-source AI technologies with the goal of ensuring AI benefits all of humanity. The video discusses the role of OpenAI and its partnership with Microsoft, as well as the governance structure that aims to democratize control over AI technologies. The script mentions 'WE THINK IT IS GREAT. ANY DEEP COMP LOOKS, RELATIONSHIP, IT IS NOT WITHOUT ITS CHALLENGES BUT IT IS REALLY, GREAT', referring to the partnership with Microsoft and the collaborative approach to AI development.

💡Regulatory capture

Regulatory capture is a form of government failure that occurs when a regulatory agency, created to act in the public interest, instead advances the commercial or political concerns of the industry it is supposed to be regulating. In the video, the concept is brought up in the context of concerns that regulations might favor larger entities and stifle smaller startups. The script includes 'THAT IS WHAT I MEANT ABOUT THE REGULATORY CAPTURE', discussing the potential issue of regulations being influenced by the industry they are meant to oversee.

💡Certification system for AI

A certification system for AI would involve a formal process for evaluating and verifying the safety, quality, and performance standards of AI technologies. The video discusses the potential for such a system, emphasizing the need for external audits and safety tests for AI models above a certain capability threshold. The script states 'I THINK THERE SOME VERSION THAT IS REALLY GOOD', referring to the idea of a certification process for advanced AI models.

Highlights

The speaker has traveled extensively, engaging with users, developers, and world leaders interested in AI, revealing a global excitement and optimism towards AI's future implications.

There is a noted intensity of excitement outside the Bay Area, suggesting a broader global enthusiasm for AI's potential.

The speaker acknowledges the presence of anxiety regarding AI, emphasizing the importance of balancing excitement with thoughtful consideration of risks.

Specific feedback from over 100 pages of notes is being used to inform changes in the development approach to AI.

The desire for global cooperation in AI development is highlighted, with world leaders encouraged to work together to ensure positive outcomes.

A 22-word statement signed by the speaker warns about the dangers of AI and emphasizes the need for global priority in mitigating risks.

The potential for AI to go wrong is recognized, but the speaker expresses confidence in the development of safety systems to prevent worst-case scenarios.

The benefits of AI are deemed tremendous, with the potential to improve education and healthcare globally, and to contribute to scientific progress.

The speaker argues against stopping AI development due to risks, advocating for global regulation instead to ensure safety.

Concerns about appearing to call for regulation in public while lobbying for less in private are addressed, with the speaker affirming a consistent stance on the need for regulation.

The idea of a certification system for AI is supported, with the notion that models above a certain capability threshold should undergo external audits and safety tests.

Google's competence in the AI field is acknowledged, and the speaker expresses a healthy respect for their focus and intensity.

The speaker discusses the relationship with Microsoft, describing it as the best major partnership they have been a part of, despite its challenges.

AI's potential to both enhance and amplify human impulses is recognized, with the speaker expressing a realistic view of technology's dual nature.

The speaker addresses concerns about biases in AI, noting that models like GPT4 have shown progress in reducing implicit biases even compared to well-trained humans.

The speaker's incentives are questioned, to which he responds that financial gain is not his primary motivation, emphasizing his desire for a meaningful impact and contribution to technological progress.

The concept of AI deciding on other AI is discussed, hinting at a future where AI systems could improve themselves and discover better architectures.

The speaker expresses optimism about the potential for international collaboration on AI, even with countries like China and Russia, despite current uncertainties.

The speaker emphasizes the importance of not placing trust in a single person or company regarding AI, advocating for a democratized approach to its governance.

Transcripts

MY GUEST NOW IS THE ONE AND ONLY PERSON WHO IS GOING TO BE

DECIDING OUR FUTURES. BRAD: I DON'T THINK SO. [APPLAUSE]

EMILY: YOU HAVE BEEN EVERYWHERE. >> THAT WAS A LONG TRIP.

EMILY: YOU WERE IN RIO, TOKYO.

WHAT SURPRISED YOU MOST?

>> A LOT. IT IS LIKE A VERY SPECIAL

EXPERIENCE TO GO TALK TO PEOPLE THAT ARE USERS, DEVELOPERS, SO

RULED LEADERS INTERESTED IN AI. YOU REALLY GET AN UNDERSTANDING

OF WHAT IS GOING ON. I THINK THE BIGGEST SURPRISE

WAS THE LEVEL OF EXCITEMENT, OPTIMISM, BELIEF IN THE FUTURE

AND WHAT THIS IS GOING TO MEAN EVERYWHERE.

I KIND OF KNEW WHAT IT WAS LIKE IN THE BAY AREA AND IT WAS MUCH

MORE INTENSE EVERYWHERE ELSE. EMILY:

MORE EXCITEMENT THAN ANXIETY?

>> ANXIETY, TOO, AS THERE SHOULD BE.

YOU HAVE TO HAVE BOTH. BUT JUST THE THOUGHTFULNESS,

THE UNDERSTANDING, THE NUANCE, THE TENSION BETWEEN THE TWO.

THAT EXISTS EVERYWHERE AND PEOPLE'S DESIRE TO REALLY

FIGURE OUT HOW TO DRIVE SOCIAL WITH THIS TECH AND WHAT IT WILL

TAKE TO COME TOGETHER AS A PLANET TO REALLY MAKE SURE WE

AVOID SOME OF THE DOWNSIDE SCENARIOS WAS QUITE

SOPHISTICATED. EMILY: HOW MIGHT YOU CHANGE YOUR

APPROACH TO THE DEVELOPMENT OF AI AS A RESULT OF WHAT YOU

LEARNED?

>> THERE'S A BUNCH OF SPECIFIC FEEDBACK, MORE THAN 100 PAGES

OF NOTES FROM MEETING WITH DEVELOPERS ABOUT COMPLAINTS

PEOPLE HAVE OR THINGS THEY WANT. EMILY:

I SAW YOU TAKING HANDWRITTEN NOTES.

>> I DO TAKE HANDWRITTEN NOTES. THERE IS ALL OF THAT AND THERE

IS SORT OF THE WAY PEOPLE WANT TO CUSTOMIZE THE TOOLS, MAKE

SURE THEIR OWN VALUES, CULTURE, HISTORY, LEG WHICH ARE

REPRESENTED IN WHAT WE HAVE TO DO TO ENABLE THAT.

THERE WILL BE A BUNCH OF SPECIFIC CHANGES WE WILL GO

MAKE.

AND THEN THE DESIRE FOR THE WORLD TO COOPERATE, LIKE THE

NUMBER OF WORLD LEADERS WHO WOULD SAY THINGS LIKE I THINK

THIS IS REALLY IMPORTANT, WE WENT TO GET AGI RIGHT, TELL ALL

OF THE OTHER WORLD LEADERS I AM IN ON IT AND WORK TOGETHER.

THEY CAME UP MAYBE EVERY TIME BUT ONE. EMILY:

YOU SIGNED A 22 WORD STATEMENT WARNING ABOUT THE DANGERS OF AI.

A REIT'S OF MITIGATING THE RISK OF EXTINCTION FROM AI SHOULD BE

A GLOBAL PRIORITY ALONGSIDE OTHER SUBTLE SKILL RISKS THAT

JUST PANDEMICS AND NUCLEAR WAR." CONNECT THE DOTS FOR US.

HOW DO WE GET FROM A COOL CHATBOT TO THE END OF HUMANITY?

>> WE ARE PLANNING NOT TO. EMILY:

THAT IS THE HOPE, BUT THERE'S ALSO THE FEAR.

>> I THINK ARE MANY WAYS IT COULD GO WRONG BUT WE WORK WITH

POWERFUL TECHNOLOGY THAT CAN BE USED IN DANGEROUS WAYS VERY

FREQUENTLY IN THE WORLD. AND I THINK WE HAVE DEVELOPED

OVER THE DECADES GOOD SAFETY SYSTEM PRACTICES IN MANY

CATEGORIES. IT IS NOT PERFECT. THINGS WILL GO WRONG.

BUT I THINK WE WILL BE ABLE TO MITIGATE SOME OF THE WORST

SCENARIOS YOU CAN IMAGINE.

BIOTERRORIST, CYBERSECURITY. MANDY MOORE WE COULD TALK ABOUT.

-- MANY MORE WE COULD TALK ABOUT. THE MAIN THING I FEEL IS

IMPORTANT THROUGHOUT THIS TECHNOLOGY IS WE ARE ON AN

EXPONENTIAL CURVE AND A RELATIVELY STEEP ONE.

HUMAN INTUITION FOR EXPONENTIAL CURVES IS REALLY THAT IN

GENERAL. IT CLEARLY WAS NOT THAT IMPORTANT IN OUR EVOLUTIONARY

HISTORY. SO I THINK WE HAVE -- GIVEN WE

HAVE THAT WEAKNESS, GET TO

REALLY PUSH OURSELVES TO SAY, OK, GPT4, HOW SURE ARE WE THE

GPT9 WON'T BE? IF THERE IS EVEN A SMALL

PERCENTAGE CHANCE IT WILL BE BAD. EMILY:

IF THERE IS THAT SMALL PERCENTAGE CHANCE, WHY KEEP

DOING THIS AT ALL? WHY NOT STOP? >> A BUNCH OF REASONS.

I THINK THE UPSIDES HERE ARE TREMENDOUS, THAT OPPORTUNITY

FOR EVERYONE ON EARTH TO HAVE A BETTER QUALITY EDUCATION THAT

BASICALLY ANYONE CAN GET TODAY, THAT SEEMS REALLY IMPORTANT.

MEDICAL CARE AND WHAT I THINK IS GOING TO HAPPEN THERE,

MAKING THAT AVAILABLE TRULY GLOBALLY. THAT IS GOING TO BE

TRANSFORMATIVE. THE SCIENTIFIC PROGRESS WE ARE

GOING TO SEE. I AM A BIG BELIEVER THAT REAL

SUSTAINABLE IMPROVEMENTS IN ALL THE OF LIFE COME FROM

SCIENTIFIC TECHNOLOGICAL PROGRESS AND I THINK WE WILL

HAVE A LOT MORE OF THAT. THERE ARE THE OBVIOUS BENEFITS.

I THINK IT WOULD BE GOOD TO END POVERTY.

BUT WE HAVE TO MANAGE THROUGH THE RISK TO GET THERE.

I ALSO THINK AT THIS POINT, GIVEN HOW MUCH PEOPLE SEE THE

ECONOMIC BENEFITS POTENTIAL, NO COMPANY WOULD STOP IT.

THE GLOBAL REGULATION, WHICH ONLY THINK SHOULD BE ON THESE

POWERFUL EXISTENTIAL RISK LEVEL SYSTEMS -- OVERREGULATION IS

HARD. YOU DON'T WANT TO OVERDO IT FOR

SURE. BUT I THINK GLOBAL REGULATION

CAN HELP MAKE IT SAFE, WHICH IS A BETTER ANSWER THAN STOPPING

IT. I DON'T THINK STOPPING WOULD WORK. EMILY:

LET'S TALK ABOUT GLOBAL REGULATION.

YOU HAVE MET WITH PRESIDENT BIDEN AND THE CEOS OF MICROSOFT

AND GOOGLE, AND YOUR CONQUER REGULATION BUT WITH CAVEATS.

THE CRITICS SAY IT SOUNDS LIKE YOU'RE SAYING, REGULATE US BUT

NOT REALLY. OR THAT YOU ARE CALLING FOR

REGULATION IN PUBLIC BUT LOBBYING FOR SOMETHING ELSE IN

PRIVATE. HOW WOULD YOU RESPOND TO THAT?

>> WE ARE PUSHING FOR IN PRIVATE, TOO.

OBVIOUSLY, WE HAVE THINGS ABOUT WAY TO DO IT THAT WILL BE

EFFECTIVE AND INEFFECTIVE. WE, FOR EXAMPLE, DON'T THINK

SMALL STARTUPS AND OPEN-SOURCE BLOWS AND HIKE OF ABILITY

THRESHOLD SHOULD BE SUBJECT TO A LOT OF REGULATION.

WE HAVE SEEN WHAT HAPPENS TO COUNTRIES THAT TRY TO OVERRATE

-- OVER REGULATE TECH.

BUT ALSO, WE THINK IT IS SUPER IMPORTANT THAT AS WE THINK

ABOUT A SYSTEM THAT COULD GET A RISK LEVEL LIKE YOU'RE TALKING

ABOUT, THAT WE HAVE A GLOBAL AND COORDINATED RESPONSE AS

POSSIBLE SO WE HAVE BEEN TALKING ABOUT THAT PUBLICLY,

PRIVATELY. I THINK IT IS IMPORTANT.

YOU COULD POINT OUT WE ARE TRYING TO DO REVELATORY CAPTURE

HERE, WHATEVER, BUT I JUST -- I THINK THAT IS SO TRANSPARENTLY

INTELLECTUAL HE DISHONEST, I DON'T EVEN KNOW HOW TO RESPOND.

EMILY: THERE IS THE SKEPTIC'S VIEW

YOUR BUILDING THESE RARE ASHLEY SHOULD SHIPS WITH REGULATORS

AND IT IS GOING TO BOX OTHER STARTUPS.

>> THAT IS WHAT I MEANT ABOUT THE REGULATORY CAPTURE.

WE ARE SAYING EXPLICITLY SHOULD NOT REGULATE SMALL STARTUPS.

IT IS A BURDEN ON THEM THAT WE DON'T WANT IN SOCIETY. EMILY:

WHAT YOU THINK ABOUT THE CERTIFICATION SYSTEM OF AI?

>> I THINK THERE SOME VERSION THAT IS REALLY GOOD.

I THINK PEOPLE TRAINING MODELS THAT ARE WAY ABOVE ANY MODEL

SCALE WE HAVE TODAY, BUT ABOVE SOME CERTAIN CAPABILITY

THRESHOLD -- I THINK YOU NEED TO GO THROUGH CERTIFICATION

PROCESS FOR THAT. I THINK THERE SHOULD BE

EXTERNAL AUDITS AND SAFETY TEST. WE DO THIS FOR LOTS OF

INDUSTRIES WHERE WE CARE ABOUT SAFETY. EMILY:

ELON MUSK WAS SCARED OF GOOGLE. IS GOOGLE STILL A THREAT?

>> GOOGLE IS UNBELIEVABLY COMPETENT AND IT SEEMS LIKE

THEY'RE FOCUSED WITH AN INTENSITY. EMILY:

SO THEY ARE STILL SCARY?

>> THEY ARE A COMPANY THAT I DON'T THINK ANYONE SHOULD EVER

WRITE OFF. EMILY: WE HAVE SEEN NEW BARBS YOU AND

ELON HAVE BEEN TRADING IN PUBLIC AND IN INTERVIEWS.

>> I DON'T REALLY -- EMILY: YOU ARE RESPONDING.

YOU ARE RESPONDING TO PEOPLE ASKING LIKE PEOPLE LIKE ME.

WHY DO YOU THINK HE IS SO FRUSTRATED OR DISAPPOINTED WITH

THE DIRECTION THAT OPENAI HAS GONE?

>> I MEAN, YOU SHOULD ASK HIM. HE CAN GIVE YOU BETTER ANSWER.

I CAN SPECULATE.

I AM HAPPY TO TALK ABOUT THIS.

I THINK HE REALLY CARES ABOUT AI SAFETY A LOT, AND I THINK

THAT IS WHERE IT IS COMING FROM. A GOOD PLACE.

WE JUST HAVE A DIFFERENCE OF OPINION ON SOME PARTS BUT WE

BOTH CARE ABOUT THAT. HE WANTS TO MAKE SURE THAT WE

THE WORLD HAVE THE MAXIMAL CHANCE. EMILY:

YOU'RE NOT WORRIED HE'S GOING TO CALL YOU OUT, CALL YOU TO

COME TO SOME CAGE MATCH IN THE VEGAS OCTAGON LIKE HE JUST DID

WITH MARK ZUCKERBERG?

>> I WILL GO WATCH IT AND ZUCK DID THAT. EMILY:

MUCH HAS BEEN MADE OF THE MICROSOFT RELATIONSHIP.

IT IS NOT JUST HIM, BUT HE HAS SAID HE IS WORRIED MICROSOFT

HAS MORE CONTROL THAN THE LEADERSHIP AT OPENAI HAS.

>> I THINK WHAT HE MEANS IS THEY COULD BREAK THE CONTRACT

AND TAKE AWAY OUR ACCESS TO THE DATA CENTER. EMILY:

AND A LOT OF MONEY THAT YOU HAVE ACCESS TO.

>> WE HAVE MONEY, IT IS THE DATA CENTER THEY OPERATE. EMILY:

HOW WOULD YOU CHARACTERIZE THE RELATIONSHIP?

>> WE THINK IT IS GREAT. ANY DEEP COMP LOOKS

RELATIONSHIP, IT IS NOT WITHOUT ITS CHALLENGES BUT IT IS REALLY

GREAT. IT IS BY FAR THE BEST MAJOR

PARTNERSHIP I HAVE EVER BEEN A PART OF.

IT IS KIND OF LIKE -- ON BOTH SIDES, IT WAS A CRAZY THING TO

JUMP INTO. SURPRISING IT WORKS THIS WELL.

BUT IF YOU LOOK AT THE RESULT, WE ARE VERY HAPPY. EMILY:

IN 2018 COME THE LAST TIME WE TALK IN PERSON, YOU TOLD ME YOU

THOUGHT AI WOULD HELP US BE OUR BEST BUT ALSO STOP OUR WORST

IMPULSES.

WHAT MAKES YOU COMPETENT ABOUT THAT BECAUSE SO MANY TIMES,

TIME AND TIME AGAIN, IF IT PSYCH TECHNOLOGY HAS ONLY

AMPLIFIED -- TECHNOLOGY HAS ONLY AMPLIFIED OUR WORST. >>

WILL DO BAD THINGS, TOO. I DON'T HAVE THIS ONLY VIEW OF

IT, I HAVE A REALISTIC VIEW. IT IS HUMAN NATURE TO TALK

ABOUT THE BAD MORE THAN THE GOOD.

I THINK YOU CAN LOOK AT OTHER TECHNOLOGIES THAT HAVE DONE A

LOT OF GOOD AND PLENTY OF HARM AND TALK 99% ABOUT THE HARM AND

1% ABOUT THE GOOD. I DID THAT, TOO.

THAT IS UNDERSTANDABLE. IN 2018, THAT WAS WAY BEFORE

THE GPT SERIES WAS A THING SO AT THAT POINT WE HAD SOME

INKLING IT WOULD GO LIKE THIS, WE CERTAINLY DID NOT KNOW

EXACTLY. BUT I THINK WHAT WE ARE HEADING

TO IS THIS PERSONAL TOOL THAT CAN HELP YOU IN WHATEVER WAY

YOU WOULD LIKE. ONE OF THE FUN PARTS OF THE

TRIP WAS HOW DIVERSE AND BROAD THE STORIES ARE OF HOW PEOPLE

ARE USING IT AT WHATEVER THEY WANT TO BE BETTER AT AND HELP

THEM.

I THINK IF YOU GO TALK TO CHATGPT USERS, WILL FIND A LOT

OF SUPPORT AND YOU CAN ALSO FIND PEOPLE WHO ARE MISUSING IT.

EMILY: THE REALITY IS YOUR BUILDING IS

ON THE BACK OF HUMAN DATA, THAT IS BIASED, RACIST, SEXIST,

EMOTIONAL, THAT IS WRONG. A LOT IS WRONG.

HOW DO YOU SAFEGUARD AGAINST THAT?

>> THERE WAS A RECENT STUDY THAT GPT4, THE MODEL THAT IS

RELEASED, IS LESS BIASED ON IMPLICIT BIAS TO TEST EVEN FOR

HUMANS WHO THINK THEY HAVE REALLY TRAINED THEMSELVES TO BE

GOOD AT THIS. IF YOU LOOK AT THE MODEL THAT

COMES OUT OF THE PRETRAINING PROCESS, THAT MODEL WILL BE

QUITE BIASED AND WILL REFLECT THE INTERNET.

BUT REINFORCEMENT LEARNING HUMAN FEEDBACK, ONE OF THE

TECHNIQUES WE USE TO ALIGN THE MODELS, WORKS QUITE WELL.

IF YOU LOOK AT THE PROGRESS FROM MODEL TO MODEL, EVEN SOME

OF OUR BIGGEST CRITICS ARE LIKE, WOW, THEY HAVE GOTTEN A

LOT OF THE BIAS OUT OF THE MODELS.

I THINK IT CAN BE A FORCE FOR REDUCING BIAS IN THE WORLD, NOT

FOR ENHANCING IT. THERE ARE QUESTIONS ABOUT WHAT

IF THE USER WANTS TO USE THE MODEL IN A BIASED WAY?

HOW MUCH CONTROL DO YOU GET A USER?

WHO DECIDES THAT LIMITS OF THE VALUE SYSTEM?

THAT WILL BE A TOUGH QUESTION FOR SOCIETY TO WRESTLE WITH.

THERE IS NOT A ONE SENTENCE BUTTONED UP ANSWER, BUT THE

TECHNOLOGY I THINK HAS GONE MUCH FURTHER THAN PEOPLE

THOUGHT IT WAS GOING TO IN TERMS OF BEING ABLE TO ALIGN

THESE MODELS TO BEHAVE IN CERTAIN WAYS. EMILY:

WE HAVE BEEN TALKING GOING BACK TO YOUR DAYS AT N.Y.C.

AND IT HAS BEEN FUN TO WATCH THAT JOURNEY.

I THINK PEOPLE REALLY WANT TO UNDERSTAND YOUR INCENTIVES AND

DON'T NECESSARILY UNDERSTAND YOUR INCENTIVES.

PEOPLE ARE PERPLEXED. THERE PERPLEXED YOU HAVE NO

EQUITY. CAN YOU EXPLAIN THAT A LITTLE?

IS THERE ANY FINANCIAL STRUCTURE WHEREBY YOU DO

BENEFIT IF OPENAI IS A BIG THING?

>> I GET WHY PEOPLE ARE PERPLEXED ABOUT THIS, AND I

HAVE WONDERED IF I SHOULD JUST TAKE ONE SHARE OF EQUITY SO I

NEVER HAVE TO ANSWER THIS QUESTION AGAIN. A FEW THINGS.

ONE, WE ARE GOVERNED BY NONPROFIT WHICH I AM A BOARD

MEMBER AND OUR BOARD NEEDS TO HAVE A MAJORITY OF

DISINTERESTED INVESTORS. LIKE, DON'T HAVE EQUITY IN THE

COMPANY. I ORIGINALLY DID NOT DO IT FOR

THAT REASON. EMILY:

BUT ARE THERE ANY FINANCIAL INCENTIVES?

LIKE ON A CERTAIN BENCHMARK?

>> NO. I HAVE A TINY BIT OF INVESTMENT

BUT IT IS IMMATERIAL. EMILY:

IF OPENAI IS MASSIVELY PROFITABLE, YOU WON'T BENEFIT

FINANCIALLY?

>> ONE OF THE TAKEAWAYS I HAVE LEARNED FROM QUESTIONS LIKE

THIS IS THIS CONCEPT OF HAVING ENOUGH MONEY IS NOT SOMETHING

THAT IS EASY TO GET ACROSS TO OTHER PEOPLE. EMILY:

IT IS HARD FOR PEOPLE TO UNDERSTAND. [APPLAUSE]

>> I HAVE ENOUGH MONEY. I'M GOING TO MAKE WAY, WAY MORE

FROM OTHER INVESTMENTS.

IF I JUST HAD TAKEN THE EQUITY, PEOPLE WOULD BE LIKE, THAT

MAKES SENSE.

THEY GOES TO THE NONPROFIT AND I TRUST THE PROPHET TO DO A

GOOD THING WITH IT, BUT I HAVE ENOUGH MONEY.

WHAT I WANT MORE OF IS AN INTERESTING LIFE IMPACT.

I STILL GET A LOT OF SELFISH BENEFIT FROM THIS LIKE, WHAT

ELSE AM I GOING TO DO WITH MY TIME? THIS IS REALLY GREAT.

I CAN'T IMAGINE A MORE INTERESTING LIFE THAN THIS ONE

AND A MORE INTERESTING THING TO WORK ON.

I GET A TON OF BENEFIT BUT, YES, SOMEHOW THIS IDEA OF

HAVING -- IT DOESN'T COMPETE FOR PEOPLE. EMILY:

IS THAT ABOUT POWER? CONTROL?

>> I WANT TO MAKE MY CONTRIBUTION BACK TO HUMAN

TECHNOLOGICAL PROGRESS. I GET TO BENEFIT FROM ALL OF

THIS STUFF THAT PEOPLE DID BEFORE.

I GET TO USE THIS IPHONE THAT I STILL MARVEL AT EVERY DAY, ALL

OF THE WORK THAT HAD TO GO INTO THAT.

THOSE PEOPLE, I DON'T KNOW WHO THEY ARE, I AM VERY GRATEFUL TO

THEM. THEY KNEW THEY WERE NEVER GOING

TO GET RECOGNITION FOR ME PERSONALLY, BUT THEY ALSO

WANTED TO DO SOMETHING TO CONTRIBUTE. AND SO DO I.

I CAN'T IMAGINE BETTER COMPENSATION OR FEELING.

IT WOULD BE MAYBE WEIRD IF I HAD NOT ALREADY MADE A BUNCH OF

MONEY AND PLANNED TO MAKE WAY MORE FROM OTHER INVESTMENTS.

I JUST DON'T THINK ABOUT IT.

-- I THINK THAT THIS WILL BE LIKE, I THINK IT WILL JUST BE

THE MOST IMPORTANT STEP YET THAT HUMANITY HAS TO GET

THROUGH WITH TECHNOLOGY. AND I CARE ABOUT THAT.

EMILY: WHAT IS ONE QUESTION PEOPLE

REALLY WISH PEOPLE LIKE ME ASKED? SAM:

THE PERSON [LAUGHTER] DRAMA OF THE DAY. EMILY:

SERIOUSLY. I UNDERSTAND YOU ARE GETTING A

LOT OF THOSE QUESTIONS. SAM:

I'M ALWAYS EXCITED TO JUST TALK ABOUT WHAT CAN HAPPEN IN THE

COMING FEW YEARS AND DECADES WITH THE TECHNOLOGY.

EMILY: SO WHAT DO WE HAVE TO DO? SAM:

TALK ABOUT ELON MUSK? [LAUGHTER] WE DON'T TALK ABOUT, IT'S

DEEPLY IN OUR NATURE TO WANT TO CREATE, TO WANT TO BE USEFUL.

TO WANT TO LIKE FEEL THE FULFILLMENT OF DOING SOMETHING

THAT MATTERS.

IF YOU TALK TO PEOPLE FROM THOUSANDS OF YEARS AGO,

HUNDREDS OF YEARS AGO, THE WORK WE DO NOW WOULD HAVE SEEMED,

YOU KNOW, UNIMAGINABLE BEST AND PROBABLY TRIVIAL.

THIS IS NOT DIRECTLY NECESSARY FOR OUR SURVIVAL IN THE SENSE

OF LIKE FOOD OR WHATEVER. THE SHIFT HAPPENS WITH EVERY

TECHNOLOGY REVOLUTION. ANY WE WORRY ABOUT THE PEOPLE

DO ON THE OTHER SIDE AND EVERY TIME WE FIND THINGS AND I

SUSPECT -- EXPECT NOT ONLY WILL THIS BE AN EXCEPTION TO THAT,

BUT THE THINGS WE FIND WILL BE BETTER, MORE INTERESTING, AND

MORE IMPACTFUL THAN EVER BEFORE. A LOT OF PEOPLE TALK ABOUT AI

AS THE LAST TECHNOLOGICAL REVOLUTION.

I SUSPECT THAT FROM THE OTHER SIDE IT WILL LOOK LIKE THE

FIRST. LIKE THE OTHER STUFF WILL BE SO SMALL IN CAN -- IN

COMPARISON.

THE TECHNOLOGICAL REVOLUTION, IT'S A CONTINUOUS ONE.

CONTINUING AND EXPONENTIAL. WHAT WILL BE ENABLED, WHAT WE

CAN'T IMAGINE ON THE OTHER SIDE, WE WILL HAVE TOO MUCH.

IF YOU WANT TO SIT AROUND AND DO NOTHING, THAT WILL BE GOOD,

TOO. EMILY: BONBONS AND PEACHES IN MY

FUTURE. SAM: I DON'T THINK THAT IS WHAT YOU

WILL WANT, BUT IT'S UP TO YOU. EMILY:

YOU TALKED ABOUT AI DECIDING OTHER AI. SAM:

THIS IS THE CLASSIC SCI-FI IDEA. THAT AT SOME POINT THESE

SYSTEMS CAN HELP ADDRESS THEMSELVES, CAN DISCOVER BETTER

ARCHITECTURES AND WRITE THEIR OWN CODE.

I THINK WE ARE ALWAYS AWAY FROM THAT THAT IT IS WORTH PAYING

ATTENTION TO. SAM: YOU DIDN'T -- EMILY: CHINA AND

RUSSIA, YOU DIDN'T GO THERE? SAM: I DID SPEAK THERE. EMILY:

VIRTUALLY? SAM: YEAH. EMILY: WHERE ARE THEY ON AI AND SHOULD

WE BE WORRIED? SAM: I DON'T HAVE A GREAT SENSE.

EMILY: DOES IT CONCERN YOU THAT WE

DON'T KNOW? SAM: YEAH, I MEAN AGAIN LIKE

ANYTHING, IMPERFECT INFORMATION CAUSES CONCERN.

I WOULD LOVE TO HAVE A BETTER SENSE.

BUT YOU KNOW, I'M OPTIMISTIC THAT WE CAN FIND SOME SORT OF

COLLABORATIVE THING AND I THINK THE THING THAT GETS SAID IN THE

U.S., THAT IT'S IMPOSSIBLE TO COOPERATE WITH CHINA, THAT IT'S

OFF THE TABLE, IT'S ASSERTED AND PEOPLE ARE TRYING TO WILL

IT INTO EXISTENCE BUT IT'S NOT CLEAR TO ME THAT THAT'S TRUE IN

FACT I SUSPECT IT'S NOT.

EMILY: I'M SO GRATEFUL YOU HAVE BEEN

AROUND THE WORLD TALKING ABOUT THIS ENTER WITH US HERE TODAY.

EVEN YOU WOULD ACKNOWLEDGE THAT YOU HAVE AN INCREDIBLE AMOUNT

OF POWER AT THIS MOMENT IN TIME. WHY SHOULD WE TRUST YOU? SAM:

YOU SHOULDN'T. LIKE YOU KNOW, I DON'T, YOU

HAVE KNOWN ME FOR A LONG TIME. I WOULD RATHER BE IN THE OFFICE

WORKING. BUT AT THIS MOMENT IN TIME

PEOPLE DESERVE BASICALLY AS MUCH TIME ASKING QUESTIONS AS

THEY WANT.

BUT MORE TO THAT, LIKE KNOW ONE PERSON SHOULD BE TRUSTED HERE.

I DON'T WANT 7 BILLION SHARES. THE BOARD OVERTIME TIME NEEDS

TO GET LIKE DEMOCRATIZED TO ALL OF HUMANITY.

THERE ARE MANY WAYS THAT COULD BE IMPLEMENTED.

BUT THE REASON FOR OUR STRUCTURE, THE REASON IT IS SO

WEIRD, THE CONSEQUENCE OF THAT READ -- WEIRDNESS IS WE THINK

THIS TECHNOLOGY, THE BENEFITS, THE ACCESS TO IT BELONGS TO THE

WHOLE.

LIKE IF THIS REALLY WORKS IT'S LIKE QUITE A POWERFUL

TECHNOLOGY AND YOU SHOULD NOT TRUST ONE COMPANY AND CERTAINLY

NOT ONE PERSON. EMILY:

ARE YOU SAYING WE SHOULD? -- SHOULDN'T TRUST OPENAI? SAM:

IF WE ARE A FEW YEARS DOWN THE ROAD AND HAVEN'T FIGURED OUT

HOW TO START DEMOCRATIZED TROLL, YOU SHOULDN'T.

BUT LIKE IF WE FIGURE OUT SOME SORT OF NEW STRUCTURE WHERE

OPENAI IS LIKE GOVERNED BY HUMANITY, AND THAT COULD HAPPEN

IN MANY WAYS, INCLUDING THE ALIGNMENT THAT WE PICK.

IT COULD MEAN ACTUAL BOARD CONTROL.

WE ARE TALKING TO A LOT OF PEOPLE ABOUT WHAT I COULD LOOK

LIKE IF WE DON'T DO THAT, I DON'T THINK LIKE JUST TRUST US

IS GOOD ENOUGH. EMILY:

WELL THANK YOU FOR EXPLAINING WHY WE SHOULD MAYBE CONSIDER

TRUSTING YOU. YOU HAVE A PLANE TO CATCH.

WE ARE SO GRATEFUL FOR YOUR TIME. SAM:

THANK YOU SO MUCH FOR HAVING ME. EMILY: THANK YOU SO MUCH. SAM:

FOR SURE.