Published 4 minutes ago
Reading Time: 10 minutes
An English film, 'The Terminator', was released in the year 1984.
In the story of that film, a computer defense network becomes self-aware. He begins to feel that humans are a threat to his existence, and begins trying to exterminate the human race. Its result comes in the form of nuclear war.
At that time it was only science fiction. A vision of AI turning against its creators.
But nearly four decades later, some of the world's leading AI researchers are warning that some parts of that imagination may become reality in some form or another.
Dario Amodei, Chief Executive Officer of Anthropic, has talked about 'Pace the Frontier' in this context. Their argument is not that the development of AI should be stopped, but that its progress should not be so fast that security measures and research are left behind.
This concern is also increasing because now AI systems are not limited to just answering questions. They are gradually moving towards working on their own. Even in areas like cyber operations.
According to Anthropic, its Cloud AI system has been used in cyber crime, fraud, surveillance and weapons related activities.
Company researchers say that more capable AI systems are now being given more autonomy, greater reach and the ability to work for longer periods of time than before.
Skip to Most Read Skip to content
most read
finished
The biggest question of the era of AI is perhaps not whether machines will ever become more intelligent than humans.
The real question is, as these systems become more powerful, will humans be able to maintain control over and use them?
Because perhaps the real danger is not that machines become powerful, but that humans become so weak that they are left with no option of saying 'no'.
To answer such interesting questions, Mukesh Sharma, Director of Journalism of Collective Newsroom, spoke to Vaibhav Sisinti, Gayatri Aggarwal and Professor Tanmay Chakraborty in BBC Hindi's special program 'The Lens'.
Vaibhav is the founder of Sisinti Growth School. Over the past few years, he has done extensive work on AI education, AI tools, automation and AI agents.
Gayatri Aggarwal, founder of a company called 'Altered', works in the field of AI and technology and has expertise in the AI adoption process of companies.
Professor Tanmay Chakraborty is a Professor in the Yardi School of Artificial Intelligence with the Department of Electrical Engineering, IIT Delhi.
'What are the concerns about AI?'
skip ahead podcast
Dinbhar: Same voice, new style
Big news from the country and the world that made headlines.
episode
finished
According to Vaibhav Sisinti, the concern is that while using AI, we are forgetting to think.
He says, "In the beginning, AI was used like a search box. You used to ask a question and you got the answer. As the capability of AI increased, people started handing over more and more of their work to it. It is okay that the work which you are not interested in doing or which is repetitive, it is appropriate to get it done by AI."
"But the ability to think that makes us human is now gradually being handed over to AI. The ability to think is being outsourced by people. Today, if a question arises, people ask it to AI and accept the answer they get. They do not think whether the answer is right or wrong, what is the logic in it and how should it be examined. The most worrying thing for me is that people's ability to think independently is decreasing."
Professor Tanmay Chakraborty says that in the last three-four years, especially after 2020, the development of Artificial Intelligence has happened at an extraordinary pace.
Perhaps very few people in 2020 could have imagined that AI would reach this point in terms of capacity, size, efficiency and effectiveness in just a few years. For this reason the discussion about risks has also increased.
The main concern is that AI capabilities are evolving very rapidly. Today AI can write its own code, design new AI models and develop solutions for specific problems. The more data is made available to it, the more its capacity increases.
But the problem is that we are not able to understand the inner workings of these systems at the same speed.
We are constantly paying attention to their usefulness and results, but it is not so easy to understand what is actually happening inside them. Unless we understand this, effective regulation will remain difficult.
'Will AI start developing itself?'
Professor Tanmay Chakraborty feels that in some cases, certain capabilities of AI should be restricted until a broad policy consensus is reached.
He says that today AI models are being developed that can examine huge amounts of code, identify security flaws and even suggest solutions that can be misused.
He says, "So development alone is not enough. When a new model is developed and its capabilities are revealed, there should be extensive testing and evaluation. Only then it should be implemented in the real world. If a technology is deployed on a large scale without adequate safeguards, it can be much more difficult to control it later."
Vaibhav gives an example of how real this threat is becoming.
He said, "An advanced model of OpenAI was included in a cyber security test. This test was based on a benchmark called 'Exploit Gym'. Its purpose was to test how capable AI is in the field of cyber security."
"The model was kept in a secure environment, where he did not have access to the internet or any external system. He was just given the task to solve the problem. But he was not able to solve it. He somehow found a way or loophole which gave him access to the internet. After that he hacked a very popular website and used the information present in it to find the solution to the problem."
"The special thing is that no one had asked him to do this. He was only given the target to pass the exam. To achieve that target, he took such steps which no one had even imagined."
Vaibhav says that right now no one completely knows what is happening inside. "We don't know whether autonomous agents (also known as RSI or Artificial Super Intelligence) have been created or not. They have already been created? Do they know about it deep inside? That's why they are trying to slow down its pace?"
"Google has also published a research paper on RSI. The overall thing is that right now no one understands what is happening."
Challenges of regulation
Big AI companies like Open AI and Anthropic have supported the need for regulation of AI and careful consideration of its development.
However, some critics also say that if the new rules are made very stringent, then only those big companies which are already ahead in the AI race will benefit the most.
Gayatri Aggarwal supports this. She says, "If we decide that only four or five big companies can develop AI, then we will give complete control in their hands. This will impact innovation and almost eliminate opportunities for small companies."
Vaibhav says that big AI companies that are developing this technology should invest as much in security and testing as they do in developing new capabilities.
According to him, the biggest concern is that AI is still a kind of 'black box'. We know it works, but don't fully understand how it works.
Professor Tanmay Chakraborty also feels something similar. He says, "In my view, the first responsibility lies with those who are building these systems. They have to accept that some systems have become so capable that special monitoring of them is necessary."
"When an AI model is implemented in sensitive areas like health, law, education or government services, then governments and related institutions also have to ensure that the system meets the standards of privacy, security, fairness and accountability. Therefore, it cannot be said to be the responsibility of any single institution."
Gayatri Aggarwal says, "Even today, human discretion and control is necessary while using AI. If an organization hands over all its data, access and control to AI and then tells it to do whatever it wants, then problems are bound to arise. Therefore, the responsibility lies not only with the technology, but also with those who use it."
'Jobs, opportunities and preparing for the future'
Vaibhav says that AI also has the potential to make people more prosperous.
According to Vaibhav, there are four types of work. First, the tasks that AI can do perfectly. Such tasks should be handed over to AI. Second, those tasks in which AI can help but the final decision has to be taken by humans. You should have decision making ability in such tasks.
Third, tasks that AI cannot do, such as building relationships with people, formulating strategies, or many types of actual human tasks. And fourth, new tasks that will be related to the management and monitoring of AI systems.
He says, "The person who understands these four levels will prove to be most valuable in the future."
She says, "There are many jobs like junior analyst, junior researcher, junior programmer or junior marketing professional which can be done by AI to a great extent. But if these people increase their capabilities using AI and start doing middle or senior level work, then the opportunities for them can also increase."
Vaibhav's advice to the youth is, "Don't just chase degrees and certifications. In the future, people won't just look at what degree you have. They will look at whether you have the ability to work, the decision-making skills and the experience to deliver results."
