00:00Okay, let's widen the debate then on this critical issue. Will AI put human race at risk as some are
00:07fearing? Are the fears of AI company researchers genuine or not? Is it time for a global safety AI network?
00:16And should there be a kill switch that can in a way effectively stop, if at all possible, some of
00:23these technologies?
00:25Those are some of the questions I want to raise. Joining me now, special guest, joined by CP Goodnani, former
00:31CEO at Tech Mahindra, co-founder and vice chairman of INOS joins me. Also being joined by Kareen Ben-Simon.
00:39She's in Tel Aviv. She's VP Global Partnerships at Dream. It's a sovereign AI and cyber unicorn valued at 3
00:47billion US dollars.
00:48And Lalitesh Katragada is also joining me, founder of Indiehood.com and an AI expert. Good to have all of
00:55you on the show. I want to go first to you in Tel Aviv, Kareen.
00:59How are you seeing this entire controversy that's now snowballing across the world, started with an anthropic researcher quitting the
01:07company, saying AI companies are gambling with our lives and pose some kind of a risk to humankind? Do you
01:15think it's an exaggerated fear?
01:18So I believe that eventually government, in order to be able to defeat this race, they have to employ and
01:28deploy also sovereign AI capabilities.
01:30You have to be able to control the data. You have to be able to own the data and the
01:36infrastructure below it. So regardless of any, let's say, innovative or attackers capabilities, you will have to make sure that
01:46you have the necessary measures, AI based, because if not, it's not like a equivalent race to be able to
01:54protect your entities and your data and your government critical infrastructure.
02:06So are you saying that governments will have the power to have some kind of a kill switch, a safety
02:12or security mechanism that can shut down an AI system or machine if there is an emergency or some danger
02:19that's seen? Is that what you're saying? Is that feasible?
02:22So I'm saying that eventually each government, when they deploy or when they use AI, they have to put the
02:31necessary measures.
02:32Each government choose on how to do it, but they will have to have the necessary measures in order to
02:38protect and to make sure that everything is in place.
02:42So I think that eventually we'll have to make sure that the government has the ownership of the data and
02:49they have also the capability to understand and to control what is happening in their AI stack.
02:56If you decide to put a kill switch based on your, let's say, internal models, that's the controls that you
03:03provide.
03:03But we want to make sure that at least we know what we have inside of it in the entire
03:08AI stack.
03:13You know, CP Gudnani, you're hearing what Kareen is saying, that governments will need to have some kind of control
03:20over data.
03:21Do you believe that's at all feasible?
03:23I've had Vineet Nayar today on my show.
03:26He says these fears in any case are exaggerated.
03:28Even when the Internet first came, there were all kinds of fears about what Internet would do to our existence.
03:34Now it's AI companies, and maybe this is a valuation game.
03:37How do you respond?
03:40So let's take a step back.
03:43The first time in the history of technology, three companies agreed with each other.
03:51Elon Musk, OpenAI, and Anthropik.
03:54So you cannot say that the incident that happened at OpenAI is not a warning signal.
04:07Number two, we all also saw the three steps that the three companies have suggested.
04:17One is on-site auditor, one is lab being coordinated, and number third, agree with China on testing and limits.
04:31Now, this kind of collaboration, more to reemphasize that AI is now running faster.
04:48You know, Anthropik founder has said AI is now generating new AI.
04:55I think we need to take it very, very seriously, and there are only two or three ways of looking
05:04at it.
05:05Number one is control the usage of AI.
05:09I mean, let's not assume that this is a machine which can be given to every individual and allowed to
05:19function on its own.
05:20Number two is control the use cases, because if you run a use case, then you are basically saying, I
05:32have defined the input and I defined the output.
05:35So I can only repeat that current, I mean, they are using the word slowdown.
05:46Now, I would say, can we hit the pause button?
05:52And till we are able to...
05:55But you see, it's very interesting.
05:58Sorry, sir, sorry to intervene.
06:02You're saying we could maybe hit the pause button, but these are the very same companies that have built these
06:07large market valuations.
06:10They've got large market share.
06:11Now, suddenly, they are recognizing the dangers and the fears of AI.
06:15I mean, they've been at this for years now.
06:18Suddenly, AI becomes so dangerous.
06:20Did they not know this earlier?
06:24As you have noticed, is that some of these events have been a wake-up call.
06:32You heard about the open AI experiment of the people who are not connected, people who are working in isolation,
06:43and the agenting AI I'm talking about.
06:46Agenting AI working in isolation is able to create fake accounts and fake narrative.
06:53I don't think the authors of this software knew about it.
07:00And if they knew about it, I would say that at least now the three of them are agreeing.
07:07They are agreeing that they have a monster in hand.
07:15I would also add to that.
07:18So, you know, Mr. Gurnani is saying there's a monster in hand.
07:24Lalit Katragada, how do you respond?
07:26Do you believe there's a monster that's been created, a Frankenstein that's been created that could devour its own?
07:31Or do you believe that's an exaggeration, that actually AI is an enabler, and we've got to look at the
07:37positives, while also being conscious of security concerns that may be raised?
07:43There's a danger now that these fears could sort of create a fear-mongering, could create a sense of hysteria
07:49once again.
07:51Yeah, I think there's a little bit of all of that going on, Rajdeep.
07:57See, if you take a step back, compared to the Internet, this is the first time in human history where
08:03we have machines that can learn, learn not just what they are taught, but learn things they have not been
08:08taught.
08:08So the risks are definitely higher.
08:11But if you look very closely, the fourth AI company in the world, where the AI system, Gemini, is run
08:20by Demis.
08:20Demis has sort of agreed, but also said that some of these threats are highly exaggerated.
08:28He believes the risk is in the sub, you know, 1% and so on.
08:32But having said that, I think there is no doubt we need to have regulation.
08:36And the first point that Karina made, I completely echo.
08:41The risks of code that you don't control, the risks of data you don't control are very high.
08:46So the Indian, I think, regulation needs to recognize that.
08:50And in the long run, maybe medium term, Indian companies, whether they are customers of the software or producers of
08:56the software,
08:57we need to have our own code and we need to have our own data.
09:01But most importantly, I think we need to focus a lot more on security than we have been.
09:06But having said that, I am an optimist.
09:08I think this AI era is going to create an era of plenty.
09:11And I think we should not lose sight of that.
09:17But, you know, Lalit, if China, U.S., which are the two big powers that are in this major AI
09:24war or AI race,
09:26and they gain enormous economic and strategic advantage out of it, can we really have a global safety regime?
09:34Yes, Karin, you want to step into that?
09:35Because, you know, China and Washington, will they agree to these red lines or will national security competition make us
09:44an AI arms control regime?
09:46I mean, who's going to really bell the cat?
09:49Which governments are going to agree to this?
09:51Will China in particular agree to have the kind of control over AI and disclose that to the world?
10:00So I think that when we had the same issue with cybersecurity, let's say the offensive side had always, was
10:09always, you know, like a very advanced, very innovative in their techniques, etc.
10:13It didn't mean that the defense or the defenders had regulations and had global collaboration and had relevant, let's say,
10:23measures, security measures to make sure that, again, we minimize the race between attacker and defenders.
10:31It's the same for AI, even when we discovered a dream, an autonomous AI cyber attack recently that targeted government
10:40entities in Asia, we were able to understand how it operated and also shared the best practices with the government
10:49customers that we have with the global ecosystem.
10:52Because, again, you have to make sure that you identify the necessary countermeasures.
10:57So it doesn't mean that even if there's, you know, like a war or, let's say, attackers on the one
11:06side, you have to make sure that you empower the defenders leveraging AI capabilities.
11:11So that's what we do also with Dream.
11:13We provide this empowerment for the cyber defenders in the cybersecurity field.
Comments