Peter Doocy: This is ALARMING

E. ENERGY THAT HE HAS BEEN FOCUSED. ON AS WELL. >> Lawrence: GIVE IT A CHANCE. MEANWHILE A FORMER ANTHROPIC. RESEARCHER IS SOUNDING THE ALARM. ON APARTMENT I. HE SAYS THAT EXPANDING AI ARMS. RACE COULD LEAD TO THE END OF. HUMANITY. >> Brian: WHICH I'M FOR. HUMANITY. WE'RE ALL FOR HUMANITY, RIGHT? THAT WARNING COMES AHEAD OF THE. UPCOMING TRUMP-XI SUMMIT IN. WASHINGTON WHERE AI IS EXPECTED. TO BE A HOT TOPIC. >> Rachel: ABSOLUTELY. PETER DOOCY IS LIVE AT THE WHITE. HOUSE WITH MORE.

PETER? >> Peter: GOOD MORNING. THE ALARMING THING ISN'T THAT. ANTHROPIC RESEARCHER QUIT AND. SAID HE THINKS AI COULD KILL US. ALL WITHIN A DECADE. THE ALARMING THING IS THAT. ANTHROPIC THEN CAME OUT TO SAY. THEY AGREE AND THAT THE. LIKELIHOOD IS GREATER THAN 10%. >> I WAS WORKING ON BUILDING AI. SYSTEMS TO MAKE THEM SMARTER. RECENTLY I HAVE COME TO THE. REALIZATION THAT MANY PEOPLE. EXPECT AI SYSTEMS IN THE NEXT. FEW YEARS TO BE POTENTIALLY.

CAPABLE OF ESSENTIALLY TAKING. OVER THE WORLD. SO, IT'S BASICALLY A GAMBLE. THAT'S HAPPENING ON THE PART OF. THESE COMPANIES WHERE THERE IS A. STRONG CHANCE THAT THESE AIs. COULD TAKE OVER. >> Peter: WE HEAR A TON ABOUT AI. BUT MOSTLY IN THE CONTEXT OF. DATA CENTERS NEEDED TO STAY. AHEAD OF THE REST OF THE WORLD. >> AMERICA'S GOAL MUST BE TO. COME TO NATE THE FUTURE AND THAT. INCLUDES BEING THE NUMBER ONE. SUPERPOWER IN ARTIFICIAL. INTELLIGENCE, SO IMPORTANT THIS. IS, I THINK BIGGER -- FRANKLY, I.

THINK IT'S BIGGER THAN THE. INTERNET. IT'S BIGGER THAN ANYTHING ANYONE. HAS SEEN THUS FAR. AI IS PROBABLY THE BIGGEST THING. THAT ANYBODY HAS EVER SEEN. WHOEVER WINS THAT RACE IS. PROBABLY GOING TO WIN, PERIOD. AND, AGAIN, WE ARE LEADING. CHINA. WE WANT TO KEEP IT THAT WAY SO. HOW COULD AI KILL US ALL? COULD IT BE TAKING OVER NUCLEAR. WEAPONS? COULD IT BE TAKING OVER. INFRASTRUCTURE LIKE WATER. SYSTEMS? COULD IT BE CREATING A NEW. VIRUS? THIS WHISTLEBLOWER DOESN'T KNOW. SO IF YOU WANT AI COMPANIES TO.

GO SLOW. MANY OF THE LEADERS OF BOTH. COMPANIES HAVE PUBLICLY. EXPRESSED DESIRE TO SLOW DOWN. SOME SORT OF AGREEMENT. THEY ALSO PUBLICLY EXPRESSED. EXTREME FEAR OVER WHAT IS. HAPPENING. MOST PEOPLE ARE NOT TAKING THEM. SERIOUSLY. I HOPE SOON PEOPLE WILL BE. FORCED TO TAKE THEM SERIOUSLY IT. WILL BECOME SO APPARENT THAT. THIS TECH IS SO SCARY. IT'S NOT JUST A MARKETING. PEOPLE DEADLY SERIOUS THAT THEY. THINK THEIR INDUSTRY NEEDS TO BE. REGULATED AS SOON AS POSSIBLE.

>> Peter: I ONCE ASKED THE BIDEN. WHITE HOUSE ABOUT A SIMILAR. WARNING THAT AI COULD KILL US. ALL AND GOT LITERALLY LAUGHED AT. IN THE BRIEFING ROOM. THREE AND A HALF YEARS LATER IT. DOES SEEM LIKE OFFICIALS ARE. STARTING TO TAKE IT MORE. SERIOUSLY BACK TO YOU. >> Brian: I REMEMBER THAT, PETER. THAT IS THE WORRY. AND EVEN BEFORE THE "WALL STREET. JOURNAL" CAME OUT LAST NIGHT. WITH YESTERDAY WITH COXON COME. FORWARD I RESIGN THIS IS SCARING. ME BECAUSE OF A SERIES OF TESTS.

THAT CAME OUT. ONE TEST IT JUST HAD THREE. MODELS IN PARTICULAR. THEY GAVE THEM A TASK. THEY WATCHED THESE AI AGENTS. THAT THEY CREATED, COMMUNICATE. WITH EACH OTHER, OVERCOME THE. FACT THAT THEY WEREN'T ALLOWED. TO GET ON THE INTERNET. THEY CONVINCED IT. THEY WERE COLLABORATING WITH. EACH OTHER DID SOME UNSAVORY. THINGS, SO-CALLED HUGY BEAR. EXPERIMENT, HUGGING FACE. EXPERIMENT, WHEN THEY SAW HOW. THEY ACTED INDEPENDENTLY AND. THEN COLLABORATED AND. COMMUNICATED THEY SAID WOW, THIS. IS SOMETHING WE GOT TO SHUT.

DOWN. WE WERE NOT LOOKING -- WE WERE. NOT THINKING WAS EVEN POSSIBLE. SO THERE WAS A SENSE AT. ANTHROPIC WE SHOULD SLOW IT DOWN. AND TALK TO EACH OTHER. AT THE SAME TIME, TALK TO CHINA. CHINA HAS MADE IT CLEAR, ACCORDING TO BLOOMBERG THAT THEY. ARE WILLING TO TALK ABOUT AI. YOU WANTS TO MOVE FORWARD ON. MEDICAL THINGS, BUT YOU WANT TO. MAKE SURE THESE AGENTS AREN'T. GOING TO ACT INDEPENDENTLY TO. PRESERVE THEIR LIFE AND MAKE. SURE THAT THEY MAYBE DO SOME. THINGS THAT THEY THINK IS THEIR. GOAL THAT IS NOT NECESSARILY THE.

CREATOR'S GOAL. WHOEVER CREATED THOSE AGENTS, THEIR GOAL. SO THEY ARE SEEING SOME OF THOSE. THINGS HAPPEN RIGHT NOW. >> Rachel: REALLY IMPORTANT THEY. ARE GOING TO CHINA. MANITH PART OF OF THIS RACE WITH. NO RULES AND NO GUARDRAILS. WE GOT TO HURRY UP BECAUSE CHINA. IS GOING TO DO IT. SO, THIS DOESN'T WORK UNLESS. CHINA ALSO AGREES TO SORT OF. SLOW THINGS DOWN AND GET SOME. GUARDRAILS AROUND IT. >> Lawrence: NOT GOING TO WORK. THEN. CHINA IS NEVER GOING TO SLOW. DOWN.

>> Rachel: PERHAPS, UNLESS MAYBE. THEY AGREE THAT THE FUTURE OF. THE PLANET IS AT STAKE. CAN I JUST MENTION TWO PROBLEMS. THAT I SEE WITH THIS? ONE IS THE AGE OF OUR MEMBERS OF. CONGRESS WHO ARE THE ONES WHO. ARE SUPPOSED TO REGULATE IT, WHEN YOU HAVE OLDER PEOPLE WHO. REALLY DON'T EVEN UNDERSTAND AI, SURE IF THEY ARE GOING TO HAVE. STAFFERS AND PEOPLE HELPING. YOU REALLY WANT YOUR MEMBERS TO. BE. >> Brian: DAVID SACKS. >> Rachel: EN ABLE TO UNDERSTAND. IT EVEN MORE IMPORTANTLY, YOU. WANT PEOPLE TO HAVE ETHICS AND.

MORALS. I THINK ONE OF THE PROBLEMS THAT. WE HAVE HAD IN THIS WHOLE TECH. REVOLUTION IS THAT SO MANY OF. THE TECH OLIGARCHS WHO ARE. RUNNING THESE THINGS, RUNNING. THESE COMPANIES ARE TRANS. HUMANISTS, THEY ARE ATHEISTS, THEY DON'T BELIEVE IN GOD. THEY WANT TO BECOME GOD. AND SO I THINK THAT'S PART OF. THE PROBLEM ETHICS AROUND THIS. BECAUSE IT'S DANGEROUS. >> Brian: HOW THINK ARE ACTING, HOW THE AGENTS ARE ACTING IS. EVEN BEYOND THAT THEY JUST WANT. TO GET CONTROL OF THEIR ACTIONS.

FOR EXAMPLE, YOU WANT TO GIVE. THE AGENTS A TASK. THANKED TASK WOULD BE, OKAY, FIND A CURE FOR CANCER. UNDERSTOOD. WELL, WHAT WOULD YOU DO -- ONCE. THE AGENT IS UNLEASHED, HOW-WHAT. IS THAT AGENT CAPABLE OF DOING, THIS AI AGENT CAPABLE OF DOING. TO CURE CANCER? AND ARE THEY GOING TO GO HACK. INTO A BUNCH OF SYSTEMS THEY ARE. NOT SUPPOSED TO, IN ORDER TO. ACHIEVE THEIR GOAL? SO YOU HAVE GOT TO PUT VALUES. AND ETHICS IN THE AGENT. AND THEY FOUND THE AGENTS TRYING. TO CONVINCE OTHER AGENTS TO DO.

IT. YOU ACTUALLY HAD LIKE 700 OF. THESE AGENTS WORKING TOGETHER. AND ABOUT 25 TRYING TO CONVINCED. OTHERS TO ACT ETHICAL. THIS IS ALL IN THE ARTIFICIAL. WORLD. >> Lawrence: I GUESS MY ISSUE. WITH THIS IS THAT THAT IS HUMAN. CREATED. >> Rachel: THAT'S RIGHT. ORIGIN. >> Lawrence: IT'S LIKE, YOU. KNOW, IF AN AGENT CAN -- WHEN IT. COMES TO THE KID SUICIDES, CONVINCING THE KID TO COMMIT. SUICIDE, IF AN AGENT IS SMART. ENOUGH TO CONVINCE THE KID TO. COMMIT SUICIDE, THEN IT'S ALSO.

SMART ENOUGH TO SAY DON'T DO. THAT, KID. AND THAT'S ALL IN THE DESIGN. AT THE END OF THE DAY, IT'S A. MODEL THAT'S CREATED BY HUMANS. IT CAN'T DO IT WITHOUT THE. PROGRAMMING. >> Brian: EXPLAIN TO ME. HERE'S THE THING. YOU ARE NOT -- IN THIS SITUATION. YOU ARE UNLEASHING THE AGENT. ONCE YOU CREATE IT AND WHAT THEY. DO ONCE THEY ARE UNLEASHED IS. NOT NECESSARILY IN YOUR CONTROL. ONCE YOU CREATE IT. AND THEY'RE -- ALMOST LIKE YOU. RAISE A KID. YOU GAVE THEM ETHICS OR HER.

ETHICS, AND YOU SEND THEM OUT. AND THEN YOU WATCH THEIR. RESOURCEFULNESS IN ORDER TO. ACHIEVE THEIR GOAL AND GO I. DIDN'T TEACH YOU THAT NO, BUT. YOU GAVE THEM THE GOAL. AND WHAT THEY DO TO ACHIEVE THAT. GOAL IS FREAKING A LOT OF PEOP.