Why 2025 will take us closer to the “Turing horizon”
- Chetan Dube has spent his career focused on the Alan Turing-inspired goal of creating machines that are indistinguishable from humans.
- As CEO of Quant, Dube oversees the development of agentic AI intended to surpass human levels of intelligence.
- Here, Dube chats with Big Think about the promise and peril of enterprise AI, including ways to ensure that the technology is safe and profitable.
A decade ago, years before GPT and LLM became household acronyms, agentic AI pioneer Chetan Dube led the creation of the conversational AI platform Amelia. Back then, Amelia’s closest competition was IBM’s Watson. Following the 2024 acquisition of Amelia by SoundHound AI, Dube made his mission the delivery of “the most human AI the enterprise world has ever known,” as founder and CEO of Quant.
Driven by his foundational interest in the work of Alan Turing at college in Delhi, Dube has kept his eyes fixed on the “Turing horizon” where “machines become indistinguishable from humans.” Noting his exceptional track record for working ahead of the game in enterprise AI, Big Think sought out a chat with Dube to explore the present and near-future of agentic AI, sound out the issues bedeviling AI’s jagged frontier, and figure out some of the ways our AI future can be both safe and extremely profitable.
Big Think: When did you first become interested in AI?
Dube: I’ve been fascinated with AI since my college days in IIT (Indian Institute of Technology Delhi), when I read Alan Turing’s thesis, which starts by saying, “I propose for you to consider the question: can machines think?” Since then, there were not many nights that I didn’t go to bed wondering about whether we could breathe life into something as inanimate as voltage levels to make them indistinguishable from human beings.
Big Think: In the 90s you left your job as an IT professor at New York University to find a way to “clone human intelligence,” and you founded AI company IPsoft (in 1998) — what drove that decision?
Dube: At New York University, a large part of our research was based on using deterministic finite-state machines to clone system engineer’s behaviors. I walked into my doctoral advisor, Professor Dennis Shasha’s office, one afternoon, convinced that given a couple of summers, we could push that from system engineer’s brains to general customer care brains. My professor, being much wiser than I, cautioned that even the father of Artificial Intelligence, John McCarthy, gave up on the problem, claiming it turned out to be much harder than anticipated. But, having the benefit of naivety and ignorance, we jumped out of NYU to drive towards that goal, where machines become indistinguishable from humans. Needless to say, it has taken more than a couple of summers. Twenty-five, to be exact — but we have finally reached that elusive Turing horizon.
Big Think: As founder/CEO of IPsoft you created Amelia in 2014 — what set Amelia apart from other digital assistants in the era before the advent of LLM?
Dube: Amelia was created with the mission of bringing “digital employees” to the world. In 2014, when it came to conversational intelligence, there were only two prominent games in town: IBM Watson and Amelia. Amelia was a pioneer in getting closest to the human-level AI. At Quant, we have finally reached our mission of delivering the most human AI the enterprise world has ever known.
Big Think: How does Quant work?
Dube: Quant is the culmination of 25 years of research to clone the output of the human brain. It leverages LLM extensively (like most other solutions do), but layers on a human language compiler, which can provide deterministic outputs (which most enterprises, such as banks need) on top of an inherently probabilistic convolutional neural network (which makes the foundation of LLMs).
A large part of our research has been to eliminate hallucinations in outcomes that agentic systems can drive.
Big Think: Some companies are still cautious about working with early-stage generative AI — what are some of the roadblocks you have encountered while developing your approach to agentic AI?
Dube: Hallucinations are the biggest impediment to widespread adoption of agentic/generative AI. In particular, financial institutions have zero tolerance for any hallucinations. A large part of our research has been to eliminate hallucinations in outcomes that agentic systems can drive.
Big Think: Why will agentic AI have a profound impact on businesses?
Dube: 2025 will be pivotal — where you will start to see widespread adoption of agentic AI in enterprises. The advancement in the technology is at a point today that agentic AI agents are able to support customers 24/7. This reduces operational costs. Simultaneously, intelligent task management systems will prioritize work increasing overall efficiency. Robotic Process Automation (RPA) will also reduce human error and increase productivity via their ability to handle repetitive workflows.
Big Think: If you succeed in creating “the world’s most qualified AI employee” it seems logical to expect a profound change in human employment — how do you address fears of job displacement?
Dube: The human neocortex is the ultimate Darwinian engine. It is in a state of continuous evolution. People who wonder about displacement of jobs may be asked what happened to the horse and carriage driver. With the advent of technology, they moved on to drive cars, engines, and planes. History will repeat itself. Technology has always been the “creator” of net new jobs. We will gain more jobs than we will displace. Necessity will become the mother of invention, as given the lack of some mundane jobs, humans will be forced to evolve to the more “creative” jobs that challenge their brains more.
The challenge is that, as opposed to the Industrial Revolution which was a brawn multiplier, the agentic revolution is a brain multiplier.
The only big fear I have is the pace of the change. [The] Industrial Revolution brought similar displacement. One man, one cart. One steam engine, a hundred carts. But the pace was limited by the production capacity of steam engines. That gave enough time for man to evolve into newer jobs that are a natural derivative of any technological revolution.
The challenge is that, as opposed to the Industrial Revolution which was a brawn multiplier, the agentic revolution is a brain multiplier. That cognitive multiplier is not 100 to 1, but infinity to 1. Besides, the speed of rollout is no longer limited by the production capacity of plants, as cognitive brains can assimilate tasks overnight, once connected to the backend systems.
Big Think: As natural language processing becomes more refined and accurate, will our daily interaction with intelligent machines become more aural and less visual?
Dube: Daily interactions will be multimodal across omnichannels. You will seamlessly switch from having a conversation on a chat to voice to shifting it to FaceTime with your digital assistant, who will have hyper-personalized knowledge about you and your requirements.
Big Think: In your ideal scenario, what does the business landscape look like after agentic AI is widespread and fully integrated?
Dube: 50-70% of routine tasks will be done by agentic AI before the end of the decade, yielding a 4.4 trillion USD of impact.
Big Think: As a leader in AI, what do you think we need to keep in mind so that the full potential of AI can be achieved?
Dube: We need to be in the business of creating safe and subservient super-intelligence. AI has to be ethical, for it to be safe. AI has to exist in the service of mankind, for it to be subservient. Ethics cannot be an afterthought, or we will be at the risk of the destiny of AI being the “final invention,” as predicted by Stephen Hawking.
We teach our kids the difference between right and wrong from birth: At Christmas time, we tell them to be good for “goodness sake,” otherwise Santa will bring them coal. We have to programmatically teach our AI systems the value system of not doing the wrong things. Imagine a world where babies were not taught the difference between good and bad. We will be at the mercy of rogue citizens creating mayhem without consequences. A definite recipe for catastrophe. The same is true if we do not teach these AI systems at inception to be safe and subservient to humans.
Big Think: Where are we now in our collective AI journey?
Dube: I stated in 2014 that the only difference between science fiction and reality is time. AI technology is so revolutionary that it will take us to that horizon which we had seen a decade ago.
