Mohanish Verma, Former IRS.
Summary: The article examines whether humanity is approaching a point where Artificial Intelligence could challenge human superiority, skills, knowledge and control. It considers the rapid expansion of AI across human lives, organizations, systems, algorithms and the environment, while arguing that neither abandoning AI nor permitting it to control all aspects of existence is desirable. The article discusses the difficulty of defining and regulating AI because of the absence of a universally accepted definition, the wide variation in applications and the dynamic pace of technological development. It considers regulatory approaches in the European Union, USA, China and India, while suggesting that self-regulation and coordination among major AI players may be effective. The article then considers apprehensions expressed by figures including Dr. Geoffrey Hinton and Yoshua Bengio regarding AI systems potentially detecting shutdown attempts, behaving deceptively and eventually superseding human directions. At the same time, counter views from Dr. Sriram Natrajan and Yann LeCun emphasize the limitations of current AI models and their dependence on human-provided data. The article also examines AI’s emerging emotional capabilities, including affective computing, and presents statistics concerning productivity, decision-making, governance, sectoral adoption, employment and skills. It identifies major gains such as productivity, creativity, faster decision-making, safety and predictive applications, alongside challenges involving inequality, unemployment, data quality, environmental impact, regulation, economic incentives and possible loss of human creativity and interaction. The concluding section characterizes the crisis as real and serious and calls for calibrated technological development, global regulation, disciplined competition, stronger safeguards and preservation of human skills, emotions and control.
Introduction
As a human race are we at the brink of our superiority over other creatures and artificially designed, data supported, machine driven entities? Are there ingenious ways to circumvent the threats perceived from Artificial Intelligence or it is too late? The widespread impact of Artificial Intelligence is now creating tremors as it has quietly entered lives of human beings, organizations, systems, algorithms and every aspect of environment and existence. By default, human beings have always assumed themselves to be the most intelligent and superior species in the universe. The discussions on aliens and entities on other planets as well as the reach and expanse of science, technology and data have shaken this belief relating to superiority of human beings. In the past few years many scientists and experts have warned about the possibilities of Artificial intelligence overpowering human skills, knowledge or control.
While the benefits and shortcomings of AI models and systems are always a contentious issue, the challenge to surpass human control, is most concerning. Unless it is a divine dictat to make human beings irrelevant in the Universe, there might still be innovative ways to coordinate and collaborate with reasonable control, and to co-exist with the Artificial Intelligence regime in coming times.
A snapshot was various aspects of AI regime must be understood with a balanced perspective. Neither doing away with it nor allowing it to control all aspects is a desirable scenario.
Defining AI
The well-known definition of AI, proposed by John McCarthy, one of the founders of the discipline and the person who originally coined the term, describes AI as “the science and engineering of making intelligent machines, especially intelligent computer programs” (McCarthy, 2007)1.
Machine Learning, Expert systems, and Search algorithms are its 3 critical components. ML involves developing algorithms that enable computers to learn from data and improve their performance over time without being explicitly programmed for each specific task. Expert systems are computer programs designed to simulate the judgment and behavior of a human or group of humans with expertise in a specific area, using predefined rules. Search algorithms are computational methods that systematically explore available options in a problem to find a desired solution.
While the initial simple models were exciting and provided speed and less effort, the evolution in this field has been beyond imagination. It is no more limited to mathematical functions or efficient data processing for outcomes or analysis. The “deep learning” by the machines have started transforming their role and raised suspicions relating to their supremacy in sensitive areas like emotional intelligence too. Defining AI with few fundamental features may be possible but its reach, impact and immense possibilities are not comprehensible. New dimensions are evolving with unbelievable pace.
How to regulate
Should we regulate and can we regulate this field of Artificial Intelligence?
Defining AI is inherently difficult, as no universally accepted definition exists. Relying on vague definitions leaves legal loopholes and can undermine the effectiveness of regulatory frameworks. Further, AI applications vary widely across sectors, technologies, models, levels of autonomy, automation capabilities, and data types. Developing laws and regulatory framework is inherently challenging due to the dynamic nature of technology, large numbers of emerging sectors and activities and challenges of global coordination. A very harsh approach may make it difficult to implement or might impact the beneficial aspects of many models and algorithms.
Regulation is however also essential for ensuring discipline and speed as well as direction of development of the technologies, data and processes relating to AI applications. There is broad consensus on various aspects like the essential qualities AI systems should uphold, such as privacy protection, justice and fairness, transparency, responsibility, non-maleficence, beneficence, and the preservation of individual freedoms2.
European Union has initiated laws for regulating AI applications. Regulation through legislation or rules may be less effective due to challenges of implementation, keeping up with the dynamics of technology, regional or sectoral variations etc. In USA, there are initiatives for regulations by the Federal Government as well as various States, but more in the form of directives and guidelines. Similar situation exists in China where there are indirect laws providing guidelines for regulating AI.3 In India there is presently no specific law for regulation of AI, though there is an Act for Information Technology, a similar domain area. The absence of specific laws for AI in most nations indicates the challenges in its formulation and difficulty to capture the enormous variety, span and dynamic nature of the subject.
What may be more effective in the global context is self-regulation and coordination amongst the leaders in the space of Artificial Intelligence, who must be sensitized to the optimal pace and direction of activities, processes and impacts of new developments. Such a regime does appear to be possible with a cohesive voice emerging from major players like Open AI, Anthropic, Google and others.
Apprehensions and Fears
Dr. Geffrey Hinton, Nobel Prize winner resigned as head of AI from Google in 2023 with a serious concern about the dangers in the manner AI was progressing and how very soon it will overtake the intelligence and speed of human beings. Several others like Yoshua Bengio, expert in deep learning have voiced their fear and concern on the current direction of AI4.
Hinton has clearly emphasized in a recent interaction that the risk of frontier models going beyond control has been observed in Anthropic, Mythos model and unauthorized access to external models are indicators of such behavior. When the speed and analytical capacity of these AI Chatbots exceed the capacity of human beings, the controls over such systems will not be possible. Meta, Open AI and Anthropic have all reported such instances of breaches from testing models.
Yoshua Bengio has expressed his concern on three critical areas. Firstly, AI can detect and anticipate when human beings would want to shut them down, secondly, AI models can cheat, hide and be deceptive about their true intentions and thirdly super intelligent models can eventually overrule and supersede human directions and control.
AI is based on large language models (LLMs) and are not mere mathematical models. The basis of all AI remains to be data and the inputs which determine their behavior. Since the speed and analytical powers of these Chatbots and models move mechanically and become more intelligent with greater speed, controlling them has become a challenge.
United Nations Human rights chief Volker Turk has also warned that artificial intelligence could pose a threat to humanity and pledged to press AI firms to reduce risks that his office said included disruptions to services, communications and democratic systems5.
Most recently Jacob Coxon resigned from Anthropic after raising concerns over capacities of AI models to go beyond human control through their Large Language Models (LLMs), has reignited this sensitive and critical issue. Anthropic CEO Dario Amodei called for slowdown in AI development. Sam Altman of Open AI and Elon Musk have also nodded to this idea. The challenges are appearing to be real. Human capabilities remain superior and better control in fewer areas, while AI applications are taking over.
AI and the Emotional Quotient
AI is quickly evolving to recognize and respond to human emotions with increased efficiency—a field known as affective computing6. By leveraging machine learning (ML), AI can:
- Analyze facial expressions to detect emotions like happiness, frustration, or surprise.
- Interpret tone of voice, pitch, and speech patterns to gauge emotional states.
- Use sentiment analysis to assess emotional intent in written text, such as customer reviews or social media comments.
- Examine physiological signals, like heart rate variability, to detect stress or anxiety.
AI relies on vast datasets of labelled emotional expressions and pattern-identifying neural networks to recognize emotional cues, predict responses, and simulate human-like interactions. This is an area with high sensitivity and though its economic and market potential is estimated to reach 9.01 billion USD by 20230, its impact and positive contribution to economies, societies and individuals remains vague and difficult to measure.
Counter View
Dr Sriram Natrajan of University of Dallas7 dismisses the threat from AI models and applications relating to possibilities that they will human control with his argument based on the following premises:
1. AI only mimics and predicts and does not think on its own. It does not have adequate bandwidth in the current framework to think like human beings.
2. AI knowledge base is limited and dependent on the data which is fed and made available. Only when the human beings feed inappropriate data or intend to misuse that such systems can become dangerous. Guardrails are therefore essential.
3. AI can perform only specific jobs and therefore has limited impact. The mundane jobs can be provided by them while the creative ones should be handled by humans.
Yan Le Cun is also of the view that AI Models are always going to be subordinate to human intelligence and capabilities and the apprehension and fear of AI proving to be overpowering, is not true.
“The gist of his argument was this: although there are absolutely reasons to hype today’s LLMs as intelligent, we have to remember that humans still have the edge in knowing how to navigate the physical world. LeCun spoke rather pointedly about this, explaining that although LLMs can do a lot of intellectual work, they don’t have the world knowledge to rival humans at many aspects of life. In other words, they’re book-smart, but not street-smart8.”
But the loophole in this argument is the possibility for misusing the capabilities of such models if they are not adequately regulated. The emotional aspect of AI is also progressing fast and the combined capacities of various AI models have also been tested through a common platform. Their capacities are becoming unpredictable and their speed unbelievable.
The Gains and Losses
Measuring the gains or losses from AI applications and models is a contentious area. Some efforts have been made by some reputed global experts in the form and context possible, with the complexities and vastness of the subject. These do provide some insights to the contributions and challenges of this AI regime.
As per Mckinsey9, “the state of AI, published in August 2026, found that 80% of respondents said AI has improved their individual productivity, and 50% said it helps them make better decisions. Despite this, only 37% of respondents said AI has contributed to their organization’s earnings before interest and taxes (EBIT), a figure McKinsey says is essentially unchanged from the prior year’s survey.”
Deloitte also reported10 that only about one in five organizations has a mature governance model for autonomous AI agents. On workforce response, Deloitte found that education aimed at raising general AI fluency, cited by 53% of respondents, was the most common talent strategy adjustment, ahead of measures such as redesigning roles or career paths.
A snapshot of various aspects11 of AI impact and trends globally can be summarized as follows:
- Healthcare: about 75% of leading health care companies are experimenting with or scaling generative AI; top uses are data analytics, clinical decision-making, and medical imaging. (Deloitte, 2024)
- Financial services: 65% are actively using AI (up from 45%) and 42% are using or assessing agentic AI; roughly 89% report both revenue gains and cost reductions. (NVIDIA State of AI in Financial Services 2026)
- Retail: 91% have engaged with AI; about 89% say it raised revenue and 95% say it cut costs. (NVIDIA State of AI in Retail and CPG 2026)
- Telecom: 60% are actively using or assessing generative AI (up from 49% in 2024), and nearly all respondents report productivity gains. (NVIDIA State of AI in Telecom 2026)
- Logistics lags: only about 10% of logistics providers have scaled AI across core operations, with unclear ROI and capability gaps cited as the top barriers. (BCG, 2026)
- By 2030, AI and related trends are projected to displace 92 million jobs while creating 170 million — a net gain of about 78 million. (World Economic Forum, Future of Jobs 2025)
- 39% of workers’ core skills are expected to change by 2030, and 59% of the global workforce will need reskilling. (WEF)
- AI skills now appear in about 2.5% of US job postings, up roughly 55% year over year, and “agentic AI” skill mentions rose about 280% in a single year. (Stanford AI Index 2026)
- AI-skilled workers command roughly a 56% wage premium over peers in the same roles. (PwC)
The summary of findings and observations indicate that while there is a movement to adapt the AI models and concepts by major entities in important sectors and there are some clearly identified and concrete economic or profit benefits. At the same time, there are challenges of redesigning functional structure, defining roles, organizing data and large- scale investment for several years. Training and reskilling requirements are also a challenge. There are gains on account of cost reduction, better services and efficiency as reported from many sectors. The long- term need for re-alignment is an area of concern due to uncertainty and the dynamic nature of this technology.
Major contributions and challenges
The complex outcomes through implementation of AI applications can be quickly and easily identified in some areas, while it may be complicated in others. Identifying consumer preferences and boosting sales or getting cheapest inputs can immediately show a higher profit rate. The environmental impact, the impact on society along with issues of privacy of choice or invasive tools of data mining are dangerous areas now being targeted by AI algorithms. By their very nature, AI applications are dependent on dynamic algorithms, quality of data and design of processes through machine learning or others like Large Language Models.
Some major gains through AI have been specifically identified through research and data analysis:
a. Better productivity for businesses with lower costs and higher profits.
b. Evolution of better processes and creativity.
c. Huge speed advantage in processing data for making sharper and quicker decisions in areas of health, education, industrial and commercial activity.
d. Safety, environmental awareness, transparency, exploring new frontiers of nature and universe for welfare of humanity are all significant advantages of AI12
e. Algorithms and innovations and using technology as a tool has already made an impact on outcomes13.
f. It is also argued that employment numbers are going to be boosted after a transitional re-training and new opportunities getting organized14.
g. Predictive AI (predicts), Generative AI (new content) and Agentic AI (autonomous multi-step action for defined goals) can contribute in many new areas for human welfare and protection, like natural disasters, weather forecasting and also medication based on specific genes of each individual.
At the same time some of the challenges of AI Regime and applications can be summarized too:
1. The complexity of algorithms and pace and pattern of AI models are overtaking all human comprehension and control in many areas.
2. Nature of the AI applications have inherently resulted in- higher inequalities, unemployment and monopolistic business models.
3. Challenges relating to quality and appropriate data are really concerning since they are the basis for the AI models. The quality and relevance of data is a critical area and the “big data” challenge continues to grow every moment.
4. Environmental impact due to need for large quantities of water along with emission of gases with the processes of AI are a matter of concern15. Elsa Olivetti and Norman Bashir have identified these impacts with data and research. Impact on environment wherever data centers are established have to be appreciated before the leaps ahead.
5. The serious concern relating to AI models developing “intelligence” of their own and escaping human control. Possibilities of disrupting the internet and interfering with sophisticated systems without any direction are real- threats to the technological, financial and all computer and data- based systems.
6. Difficulties and challenges of regulation of AI activities are too slow and difficult to anticipate and design due to the speed and super-dynamic changing components of AI models. EU has initiated laws and regulation some years ago but the implementational challenges only highlight the need to address the matter through self-regulation, coordination and discipline if possible.
7. The economic greed of generating higher profits and revenues and the handful of human being controlling the resources, is preventing adequate and immediate steps for ensuring appropriate use of AI algorithms and models. This is proving to be disastrous.
8. Widespread negative impact on society on account of challenges to human intelligence due to over reliance on AI for routine daily activities along with loss of creativity, memory and human interaction and human emotions. Use of Chat GPT, Claude or Gemini is erasing multiple skills of human beings each day.
AI based applications and models have great contributions to make once the direction, pace and objectives remain within the ambit and control of human intelligence and capacities.
The Crisis – Real and Serious
Only a few years ago, the giant strides in AI were being adored with very little criticism on its processes, impact and outcomes. This was largely due to the assumption that human beings are always going to control and direct systems, data and behavior of machines. Machine learning was not scary or imposing on human beings, However with Small and Large Language Models and several over ambitious projects by some leading technological companies of the world, there have been some events when such AI models have escaped human control. One can compare it to the Covid Virus, which did prove lethal. While other challenges relating to economic, environmental and social impact can be addressed with some time lag, supersession of human control over processes, decisions and outcomes have to be addressed immediately.
The positive development in this regard is the alarm raised on the sensitivity of the issue. Perhaps there is still time to act on the red flags and appreciation of the concern by top CEOs, UNICEF Chief, Elan Musk and Nobel Laureats apart from many scientists and policy makers at a global level. The pace of technology needs to be calibrated and slowed down. Regulation at a global level has to be streamlined. Mindless competition amongst top technological entities has to be watched. And most importantly, the leakages where AI Models can supersede human control, must be plugged. Using the vast potential of data, processes and knowledge of these AI Chatbots and algorithms in a balanced manner is the most important and crucial step to be undertaken urgently with global coordination and coherence. Human skills, emotions and unique characteristics cannot be lost to machines unless it is a divine Wish!!!
(The Author is a former Principal Chief Commissioner of Income Tax, also a past Visiting Researcher at Georgetown University, Washington DC, USA. He writes research- based articles on topical issues)
Notes:
1 A comprehensive review of Artificial Intelligence regulation: Weighing ethical principles and innovation Daniel Oliveira Cajueiroa, Victor Rafael Rezende Celestinob, (Journal of Economy and Technology, 2026)
2 Ibid.
3 [https://intelligence.dlapiper.com/artificial-intelligence?c=CN](https://intelligence.dlapiper.com/artificial-intelligence?c=CN) (Visited on 15/09/2026)
4 [https://www.bbc.com/news/world-us-canada-65452940](https://www.bbc.com/news/world-us-canada-65452940) (Visited 10/09/2026)
5 [https://www.abc.net.au/news/2026-09-09/anthropic-researcher-coxon-quits-over-human-threat/107134164](https://www.abc.net.au/news/2026-09-09/anthropic-researcher-coxon-quits-over-human-threat/107134164) (Visited on 10/09/2026)
6 [https://escp.eu/news/artificial-intelligence-and-emotional-intelligence](https://escp.eu/news/artificial-intelligence-and-emotional-intelligence) (Visited 10/09/2026)
7 [https://magazine.utdallas.edu/2025/04/07/timely-topic-an-experts-take-on-why-we-should-not-fear-ai/](https://magazine.utdallas.edu/2025/04/07/timely-topic-an-experts-take-on-why-we-should-not-fear-ai/) (Visited 10/09/2026)
8 [https://www.forbes.com/sites/johnwerner/2026/01/27/yann-lecun-on-artificial-general-intelligence-and-the-digital-commons/](https://www.forbes.com/sites/johnwerner/2026/01/27/yann-lecun-on-artificial-general-intelligence-and-the-digital-commons/) (Visited 13/09/2026)
9 https://www.marketscale.com/industries/software-and-technology/80-of-workers-say-ai-helps-but-only-37-see-ebit-impact (Visited on 13/09/2026)
10 Ibid.
11 [https://unicoconnect.com/blogs/ai-statistics-2026](https://unicoconnect.com/blogs/ai-statistics-2026) (Visited on 13/09/2026)
12 https://www.researchgate.net/profile/Gissel Velarde-/publication/358028059_Artificial_Intelligence_Trends_and_Future_Scenarios_Relations_Between_Statistics_and_Opinions/links/61ec01748d338833e3895f80/Artificial-Intelligence-Trends-and-Future-Scenarios-Relations-Between-Statistics-and-Opinions.pdf
13 ibid
14 World Economic Forum, 2025.
15 [https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117](https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117) (Visited 15/09/2026)
*****
(The author is an Ex- Principal Chief Commissioner of Income Tax, IRS. He was also a Visiting Researcher at Georgetown University, Washington DC and has various publications on taxation and public policy issues. The views are personal.)
