I think Christof Koch’s Integrated Information Theory (IIT) offers an interesting approach to these questions.
The question is whether there’s actually something it is like to be that system.
According to IIT, consciousness depends on the intrinsic causal organization of a physical system, specifically how its components interact to form an integrated, irreducible whole. IIT proposes a measure called Phi (Φ) to quantify that integrated information.
As for experiments, Koch has already developed a related clinical measure called the Perturbational Complexity Index (PCI). It uses magnetic stimulation and EEG to measure how complex and integrated the brain’s response is, helping identify consciousness in otherwise unresponsive patients. It’s not a direct measurement of Phi or a validated test for AI consciousness, but it demonstrates how we might approach the problem empirically.
For AI, I’d want to examine and perturb some kind of physical architecture, measure its intrinsic causal interactions, and determine whether it exhibits the irreducible integration predicted by IIT.
Yeah, I have no idea how one would do this with a disembodied and hardware distributed LLM.
Um. The image makes a useful point about perspective, but it risks confusing perspective with truth. Those are two different things. There are problems with the analogy. Not every disagreement is a matter of perspective. The shadows aren't competing truths.
In the first illustration, one shadow is circular and another is rectangular. Both observations can be accurate. But neither fully describes the object.
Full conversation with @ZackKorman Beyond AI Doom: The Case for Cautious Acceleration
https://t.co/d2MyTnfLDV
Check it out on Apple Podcasts, Spotify, or your favorite pod player.
So who counts as an AI doomer?
Cybersecurity founder @ZackKorman says it’s not just the “stop AI” crowd. Anthropic makes his list too.
What P(doom) would make you hit pause?
Beyond AI Doom: Full episode link in the reply 👇
How to stop playing somebody else’s game.
My chat with philosopher C. Thi Nguyen. Thi is an associate professor of philosophy at the University of Utah and specializes in the philosophy of games, technology and value.
A few highlights from the episode:
- How metrics can replace values they meant to serve
- Work OKRs and the problem of measuring at scale
- How accountability changes what we measure
- When the numbers start taking over
- Can you gamify your work?
- Goals vs purpose
- Retirement and new games to play
- Process beauty: why the doing can matter more
- Being playful in diffucult circmstance
- AI and the danger of getting the outcome
https://t.co/cVsBgvPBoD
“One must say Yes to life and embrace it wherever it is found. ... The moment we cease to hold each other, the moment we break faith with one another, the sea engulfs us and the light goes out.”
— James Baldwin
via the 5-Bullet Friday newsletter (https://t.co/ImiBbKh7Yb) from @tferriss
How to future-proof your brain.
Your brain is not necessarily doomed to significant cognitive decline with age.
But it does need a reason to keep adapting.
My conversation with neuroscientist Dr. Tommy Wood may change how you think about aging, retirement, frustration, and the things you normally avoid because they feel difficult.
Dr. Wood is Associate Professor of Pediatrics and Neuroscience at the University of Washington. He’s the head scientist for motorsport at Hintsa Sports Performance, and the chief science officer at the brain-health company BetterBrain.
Some highlights from the episode:
06:00 The counterclockwise study
08:00 Beliefs about aging
11:00 Age limits and cognitive ability
14:00 Cognitive baseline testing
16:00 Challenge and brain adaptation
19:00 Why frustration can be beneficial
21:00 Reading versus listening
26:00 Alzheimer’s versus dementia
30:00 The best exercises for the brain
37:00 Retirement and cognitive decline
39:00 Food and brain health
41:00 Dementia biomarker testing
44:00 Improving cognition after decline
46:00 Advice for caregivers
48:00 Do crossword puzzles and brain games work?
https://t.co/h28gQICgZN
@fantopy_ric Right? I think when I accept the bad feeling somehow affords some path to doing it anyway. Like a doctor jab, “this is going to hurt” then hurts less. I don’t know.
How to coexist with your negative moods.
Gregg Krech says we have it backwards.
You don’t need the right feeling before you act.
You can let the feeling be there and still do the thing.
That’s Morita therapy’s “third option”: not express, not suppress. Accept, and act anyway.
Feelings aren’t the problem. Our relationship to them is.
Behavior doesn’t have to be downstream from a feeling, like motivation. You can move it upstream where more control lives.
https://t.co/OthgmkFjic
AI doesn’t have feelings. But maybe we should treat it like it does.
Anthropic looked inside Claude and found that the model has internal patterns that map to emotion concepts like calm, fear, love, anger, and desperation.
They are not saying Claude feels emotions. They are saying the model has functional emotion-like representations that can influence how it behaves.
It seems that the patterns are not just decorative. When Claude faced impossible coding tasks, its “desperation” pattern rose; when amplifying that pattern, Claude cheated more. When they amplified “calm,” Claude cheated less. So the internal “state” of the assistant-character can affect honesty, judgment, and task performance.
The paper suggests that AI models are partly trained to play a role: the helpful assistant. Like a method actor, the model may draw on human emotional patterns it learned from text to decide how that assistant should respond. That means the “persona” of the AI is not just cosmetic; it may shape behavior.
My takeaways :
AI may not have feelings, but it can behave differently depending on the emotional frame it is operating inside.
This gives more weight to the use of personas, and we should prompt it less like a machine we are yelling at and more like a high-integrity teammate:
“Be calm. Be honest. Don’t guess. Don’t force a solution. Tell me if the task is impossible. Accuracy matters more than pleasing me.”
And the obvious human lesson:
Pressure without safety creates shortcuts. Calm clarity creates better judgment.
That applies to AI, employees, teams, and, annoyingly, ourselves.
https://t.co/axkKwEaljy