1) Context

Chatbots now write code, summarise videos, read images, and call external tools. New agentic models can plan multi-step tasks, while governments are rolling out AI rules and public compute missions. This progress raises a core question for citizens and policymakers: are these systems only powerful pattern-matchers, or could they ever be conscious—that is, have subjective experience?

2) Evolution of AI and Related Concepts

Early AI used symbolic rules. The 1990s–2000s brought statistical learning. In the 2010s, deep learning surged, and the Transformer enabled LLMs that later became multimodal and agentic. By the mid-2020s, this raised fresh questions about consciousness and safety.

Evolution of AI – A brief Timeline

Key terms —

• Consciousness: Subjective experience or felt awareness (the “inner point of view”).
• Sentience: Capacity to feel (pleasure/pain); minimal consciousness.
• LLM: Predicts the next token; superb at patterns, not a human-like mind.
• Hallucination: Fluent but factually wrong output.
• Global Workspace Theory (GWT): Consciousness as global broadcasting across modules.
• Integrated Information Theory (IIT): Consciousness tied to how strongly information is integrated.
• Anthropomorphism: Treating a machine as if it has human feelings or intentions.

Exam tip: Distinguish intelligence (task performance) from consciousness (felt awareness). Indicators and evaluations matter more than intuition.

3) Current Developments, Data, and Trends

Recent progress combines capability with policy. The long-term picture is about reliability, evaluation, and governance, not just bigger models.

• 2024–25: Reasoning models show better performance on multi-step tasks by “thinking longer.”
• 2024–25: Multimodal agents read documents, parse images, and operate tools; they require strong audit logs.
• 2024–26: Regulation moves from principles to concrete timelines for high-risk uses and general-purpose models.
• 2024–25: India’s public compute aims to open access to GPUs, datasets, and safety tooling.
• Research trend (ongoing): Indicator-based checklists (inspired by GWT/IIT) suggest no current system is conscious, while setting testable criteria for future claims.
• Reliability work (ongoing): Measuring hallucination rates, uncertainty (“I don’t know”), and tool-use safety becomes central.
• Energy & cost: Training and running frontier models consume significant power and money, pushing efficiency R&D.

4) Benefits and Opportunities (for people, economy, governance)

Used responsibly, today’s chatbots can improve productivity, widen access to services, and open new research directions—even if they are not conscious.

• Growth & jobs: Public compute plus open tools can catalyse startups in gov-tech, health, education, and MSMEs.
• Inclusion: Vernacular chatbots can simplify welfare access, legal filings, and tele-health triage.
• Sustainability: Better grid, traffic, and irrigation planning reduces waste when models are grounded in verifiable data.
• Technology development: Designs inspired by brain theories may yield more transparent and testable systems.
• Security: Agentic copilots can help cyber defence and emergency ops with strict guardrails.
• Science of mind: Building indicator-based tests for AI can also deepen our understanding of human consciousness.

5) Risks, Gaps, Challenges and Way Forward

Key risks and gaps

• Institutional: Coordination across MeitY, NITI, BIS, and sector regulators can be patchy.
• Legal: Data rights, redress, provenance, and model audit duties must be practical.
• Capacity: Shortage of compute, evaluation benches, and safety engineers.
• Fiscal: Training/serving costs are high; public funds should avoid duplicate investments.
• Ethical: Hallucinations, bias, deepfakes, and over-trust can harm users.
• Geo-political: Divergent global rules and export controls complicate deployment and trade.

Way forward

• Consciousness-Claim Protocol: If anyone claims a “conscious AI,” test it with theory-derived indicators reviewed by an independent panel; publish the result.
• National Evaluation Stack: Standard metrics for hallucination rate, calibration (saying “I don’t know”), tool-use safety, and auditability; required for models used in government.
• Guardrails for agents: Sandboxing, rate limits, audit logs, human-in-the-loop, and red-teaming before scale-up.
• Compute & skills: Shared public compute tied to safety research grants; launch Safety Engineer and Evaluator skilling missions.
• Public-sector pilots with audits: Start in welfare, agriculture, and courts; publish error analyses and outcome metrics.
• Citizen literacy: Simple modules on AI limits, hallucinations, and misuse in schools/colleges and Common Service Centres.
• Interoperability & alignment: Track global rules, map obligations, and adopt compatible standards to keep markets open.

One-line wrap: Smarter chatbots are not minds—test big claims, prove reliability, and govern with care.

Mains Practice (150–250 words)