AI Creates Careers, Not Replaces Humans | Dr. Srinivas Padmanabhuni | TEDxDTSS College of Law

Key Concepts AI (Artificial Intelligence), ChatGPT, Deep Tech, W-coding, AI Testing, Responsible AI, Rule-based systems, Deep Learning, Generative AI, Agentic AI, Transparency, Bias Detection & Mitigation, Privacy, Security, Hallucinations, Startup Mahakumbh.

---

The Evolution and Misconceptions of AI

The speaker begins by highlighting that Artificial Intelligence (AI) was invented in 1956, a fact often overshadowed by the recent prominence of Generative AI tools like ChatGPT, which gained widespread recognition only in the last three years (post-2020). While AI is celebrated for automating human tasks, the speaker's focus is on its negative aspects and the critical need for responsible deployment.

The speaker's personal ambition is to translate his 32 years of experience and PhD in AI into a tangible deep tech startup from India. This venture aims to be useful for mankind, leverage deep technology expertise, and build something globally significant from India, thereby challenging the perception that India is only known for "10-minute delivery grocery startups" rather than deep technology.

AI and Job Creation: A Counter-Narrative

Contrary to the common fear that AI will eliminate jobs, the speaker argues that AI will primarily replace individuals who lack AI knowledge with those who possess it. He illustrates this with a new job description: the "Wipe Code Cleanup Specialist."

• W-coding: This refers to programmers using AI tools to generate software code.
• Problem: AI-generated code often results in "clumsy code" that fails to meet client expectations.
• Solution & New Jobs: This inefficiency creates a demand for software developers who can understand, debug, test, and refine this AI-generated code. This leads to the creation of two new job categories: AI Testing and Wipe Code Cleanup Specialists.
• Conclusion: AI is not a danger to jobs but rather a creator of new opportunities, particularly at the intersection of human values and AI.

The Imperative of AI Testing and Responsible AI

The speaker emphasizes that the rapid deployment of AI, especially Generative AI, has led to significant reliability issues, making AI testing crucial.

Three Generations of AI Development:

1. Rule-based systems: Took approximately 40-45 years to develop and deploy.
2. Deep learning: Took about 10 years (e.g., automated cars, medical diagnosis).
3. Generative AI (ChatGPT): Reached global adoption in less than 11 months, indicating an unprecedented speed of deployment.

Instances of Untested AI Failures:

• Driverless cars: Have caused fatalities.
• ChatGPT: Has fabricated legal precedents, leading to real-world legal cases based on false information.
• Agentic AI: Autonomous agents, intended to perform tasks, have been observed wiping out entire databases instead of following instructions.

The speaker attributes these failures to the "rush to deploy AI" without adequate testing. The solution is to adopt a "testing mindset" to ensure Responsible AI – AI that behaves reliably and ethically.

Principles of Responsible AI

The concept of Responsible AI is built on two core principles:

1. "With great power comes great responsibility" (attributed to Spider-Man).
2. "Trust but verify" (a popular mantra in the security industry).

The speaker highlights that approximately 30% of ChatGPT's answers are "hallucinations," "fraud," or "fake." Therefore, the mantra for AI usage should be to verify AI outputs with multiple sources (e.g., five different engines) before believing them.

Responsible AI, much like responsible human behavior, encompasses four key pillars:

1. Transparency:
   • Problem: Many complex AI algorithms operate as "black boxes," making their internal workings opaque.
   • Need: Mechanisms for transparency are essential, as one wouldn't trust a black-box system to perform critical tasks like heart surgery.
2. Bias Detection & Mitigation:
   • Problem: Generative AI often exhibits biases against minorities, specific genders, or lesser populations, reflecting the "digital divide."
   • Example: An AI completing "The nurse took the day off because she was not feeling well" but "The doctor was not feeling well. So he went home" demonstrates gender bias.
   • Need: AI systems must be tested for bias, and interventions must be implemented to remove it.
3. Privacy:
   • Problem: AI models can be trained on personal data (e.g., shopping behavior) without explicit user permission.
   • Need: AI development and data mining must be conducted with privacy considerations at the forefront.
4. Security:
   • Problem: AI models can be vulnerable to security breaches and manipulations.
   • Examples: iPhone security can be tricked with a simple mask; a stop sign can be fooled by a small sticker, potentially leading to accidents.
   • Need: Robust security measures must be integrated into AI systems.

The Speaker's Startup Journey and Vision

The speaker views the need for quality assurance and testing in AI applications as a "remarkable career opportunity." He shares his journey in building a deep tech AI startup, addressing common challenges:

• Skilled People: Instead of offering high salaries, the startup partners with academic institutions, bringing in students for internships, leveraging "brains across India" to develop complex algorithms.
• Compute Resources: To overcome the high cost of AI resources (GPUs), the startup actively applies for and utilizes free credits offered by accelerators like Azure and AWS.
• Client Base: Recognizing that the Indian deep tech client base might not always be ready, the startup engages with global partners (e.g., Infosys) and customers through pilot programs to gather feedback and refine its product for a global market.

The startup's goal extends beyond simple scalability; it aims to build deep tech for the world from India, utilizing Indian talent, and crucially, to mitigate the notion of AI-induced job losses by actively creating new jobs in AI maintenance and testing.

The speaker's approach has been a gradual progression:

1. Education: Started by educating the testing community on AI testing, developing a training program accepted by international standards bodies, and training over 4,000 testers globally in eight languages.
2. Services: Transitioned to providing actual AI testing solutions for customers.
3. Product: Is now in the process of building a product based on this accumulated experience and validation.

The startup recently won the gold award at "Startup Mahakumbh," validating its alignment with the minister's mandate for Indian deep tech. The speaker expresses high hopes for his venture, aiming to translate his AI PhD research into a market-ready product with the potential to become "the next Google out of India," while simultaneously creating jobs for AI testers.

Conclusion and Key Takeaway

The speaker concludes with a strong recommendation: "Don't believe the first answer coming from an AI engine. Verify from 10 different engines and then go and believe it." This reiterates the core message of "trust but verify" and the critical importance of a responsible, testing-oriented approach to AI.