Google DeepMind CEO Demis Hassabis has said that AI systems may develop a degree of self-awareness implicitly, even though they are not explicitly built to have it, adding that such a development is “possible.”
Google DeepMind · Demis Hassabis
Could AI develop self-awareness on its own?
Demis Hassabis says self-awareness may not be designed into AI systems but could emerge implicitly as models grow more capable. He calls this “possible,” not proven, and offered no timeline, test, or specific model as evidence.
Two ways a property can appear
Explicitly designed
Engineers deliberately build a feature into the system, specify it, and test for it directly.
Implicitly emerging
A property arises on its own as models scale, without being targeted. Hassabis says self-awareness “may happen implicitly.”
Emergence of untargeted abilities has been discussed for large models. Applying it to self-awareness is far more contested.
The claim as a chain of steps
→
Step 2
Capabilities emerge
→
Step 3
Self-awareness “possible”
→
Step 4
Evidence still missing
Key takeaway
A hedged hypothesis, not an established finding.
Hedged language
Words like “possible” and “might” signal uncertainty, not a confident prediction.
No observable test
It is unclear what behavior would count as evidence of self-awareness.
Hard to separate
Self-referential language could come from sophisticated pattern-matching rather than genuine self-awareness.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...