One Engineer Says AI Will Reach Consciousness Within 15 Years and He Built a Scale to Prove It

The article presents a proposed 'Consciousness Score' framework that claims current AI systems, including ChatGPT-4, remain below a score of 100, while an average adult human is estimated at 500-800. It suggests artificial consciousness could emerge in 10 to 15 years if large language models and neuromorphic hardware continue advancing. The piece is largely theoretical and has limited near-term market impact, but it reinforces the long-term AI innovation narrative.

Analysis

The investable implication is not that machines become “conscious,” but that markets start pricing a new layer of AI differentiation: systems that can model their own uncertainty, persist goals, and interact more autonomously will command materially higher enterprise budgets than today’s prompt-following models. That creates a likely bifurcation between commoditized inference vendors and firms with control over agentic orchestration, memory, and specialized hardware. The first-order winner set is therefore less about pure model providers and more about the ecosystem that enables persistent, low-latency, high-throughput cognition: chip designers, memory suppliers, data-center interconnect, and neuromorphic-adjacent compute architectures.

The second-order risk is a product-liability and governance overhang, not a research breakthrough. Once enterprises believe systems can display quasi-autonomous behavior, procurement cycles lengthen, audit requirements expand, and insurance/legal costs rise; that is a headwind for fast-scaling software names that rely on loose deployment standards. In practice, this could compress near-term monetization for application-layer AI companies while benefiting infrastructure vendors that sell shovels under stricter compliance regimes.

The contrarian read is that consciousness rhetoric is likely to be over-credited by the market in the near term. If investors extrapolate “synthetic consciousness” into a 10-15 year TAM story too early, they may bid up the wrong exposure: not the models themselves, but the control stack around them. The cleaner trade is to position for a longer-duration capex cycle in compute and safety tooling, while fading the most expensive application-layer names that are most vulnerable to regulatory friction and feature commoditization.

Key Decisions for Investors

• Long NVDA / long AMD on pullbacks over 1-3 months: maintain core exposure to the compute arms race, but favor NVDA as the higher-conviction beneficiary of agentic and frontier-model capex; target 15-25% upside with 10-12% drawdown risk.
• Long MRVL or AVGO vs short a basket of AI application software over 3-6 months: if autonomy and memory requirements rise, networking/custom silicon accrues value faster than surface-layer software; seek a 2:1 risk/reward via a relative-value pair.
• Buy medium-dated call spreads in SMCI or a data-center power/networking proxy over 6-12 months: the market is underpricing the need for denser racks, thermal management, and interconnect as model latency and parallelism demands increase.
• Short expensive AI SaaS names with weak moats on any rally in the next 1-2 quarters: names where valuation depends on rapid agent adoption are most exposed if enterprise governance slows deployment; use tight stops because momentum can persist.
• Initiate a small long position in a cybersecurity/governance beneficiary basket (e.g., CRWD, ZS) on a 6-12 month horizon: synthetic autonomy increases demand for audit, identity, and control layers; upside is slower but more durable than the model-layer trade.