Thinking Machines: AI & Philosophy

For years, digital mental health tried to take gold-standard therapies and copy them into apps. Dr. Jessica Flannery was part of that wave, and she now thinks it was the wrong goal. An app isn’t a therapist in the room, and judging AI against a human therapist means we keep finding the ways it falls short instead of asking what it can do that therapy can’t.

Dr. Jessica Flannery is a clinical psychologist and developmental neuroscientist who has spent her career bringing clinical science into commercial products. She built the clinical function at digital mental health companies like Limbix, which made an app for teenagers and young adults with depression. At Akili, she worked on an FDA-cleared video game to improve attention in adults with ADHD. And at Aura, she studied how kids use AI.

We discuss:

• Why copying gold-standard therapies into apps missed the mark
• What a working alliance means when one side is an AI, and why people believe an AI likes them
• The difference between a person who could judge you and chooses not to, and an AI
• Stepped care, and where AI fits between doing nothing and weekly therapy
• What happens when you ban a technology, from TikTok to AI therapy
• The difference between therapy and being therapeutic
• What makes a therapist effective, and what congruence could mean for an AI
• Why the right amount of engagement depends on what someone needs
• Whether AI is a new wave of psychotherapy or a new way to deliver it
• What Jessica would not want AI to do in mental health

Should Your Doctor’s Appointment Be Recorded? Psychiatry, AI scribes, and more with Dr. Arpan Parikh

Three years ago, if a patient had asked to record their appointment, Dr. Arpan Parikh says he would have had an allergic reaction. Today, recording is on the first page of his consent form.

Ambient AI scribes have been adopted faster than almost any tool in medicine, so doctors now routinely record their patients. Arpan argues that patients should be able to ask for the same thing back, and that a recording protects a doctor doing a good job as much as it protects the patient.

My guest is Dr. Arpan Parikh, co-founder and CEO of Kin, an app patients bring to their doctor’s appointments to record the visit and keep track of what was said. Arpan is a psychiatrist, double board certified in general psychiatry and addiction medicine. He was most recently Chief Medical Officer at SOL Mental Health, and he still sees patients.

We discuss:

• Why Arpan went from resisting patient recordings to recording his own visits
• What clinicians might lose when an AI writes their notes for them
• Why psychiatry is probably the specialty most resistant to change, and why that shouldn’t stop progress
• Why he compares recording doctors to putting body cams on police officers
• His view that the companies collecting our health data are now mostly collecting AI slop
• Undiagnosing: patients whose bipolar disorder or schizophrenia turned out to be substance use, and whether psychiatric diagnosis survives once we know much more about each patient
• AI as a new step in care that frees clinicians for the patients who need them
• What Arpan would not want AI to do in mental health

(Cross-posted from Fit Check) 

The mental health field is facing an unexpected disruption: AI is already the largest provider of mental health support in the US, with hundreds of millions using tools like ChatGPT for emotional guidance—whether the technology was designed for it or not.

In this AMA, I sit down with Dr. Rachel Wood (licensed counselor with a PhD in cyberpsychology and founder of the AI Mental Health Collective) and Megan Cornish (therapist and community organizer) to take hard questions from mental health professionals about what's actually happening—and what should happen—as AI enters therapeutic spaces.

We tackle the uncomfortable questions: Should AI therapy be regulated? What happens in the brain during AI conversations versus human ones? How do we prevent AI from replacing human connection rather than supporting it? And critically—what responsibility do AI companies have when people turn to them who need something AI can't offer?

This isn't about whether AI should be used for mental health support. That ship has sailed. This is about understanding the technology, acknowledging the massive access gap in mental healthcare (one therapist for every 1,500+ people in need), and figuring out how to build AI that strengthens human relationships rather than replacing them.

We discuss:

• The neuroscience of attunement and whether AI can replicate therapeutic connection
• How language models actually work (and why they're more than "just talking")
• Three core concerns: autonomy, competency, and relatedness
• Why ChatGPT exemptions in regulation miss the point entirely
• The difference between AI as replacement versus AI as gateway
• What accountability could actually look like in this space

Whether you're a therapist worried about the impact of AI within your profession, a technologist building in this space, or simply someone trying to understand where we're headed—this conversation matters.

For the full discussion and ongoing conversations, check out the Fit Check community.

“Everyone should go to therapy.” It’s a common statement, but much harder to achieve than it looks. There’s only one therapist for every 10K would-be clients, and the gap is only growing every year.

That gap is what my latest guest - Dr. Derrick Hull - has spent his career trying to fix. Now serving as the founding Clinical Lead at Slingshot AI, Derrick previously led Clinical R&D at Talkspace, where he was instrumental in popularizing text-message therapy, a previously controversial modality now recognized as critical to increasing accessibility to mental health care. Derrick has had a fascinating career, serving as a practicing psychologist, a clinical advisor and researcher for top startups, and a highly accomplished academic with publications in Nature, the National Academy of Sciences, and Oxford University Press, with his research supported via R01 grants through NIH. 

It’s been amazing getting to work with Derrick at Slingshot, and I’m so excited to share just a snapshot of the wisdom and insight I collect from him everyday.

 In our latest conversation, we touch on...

• The ethical problems (and opportunities) of AI therapy
• The role of empathy in practicing psychology
• Why being a therapist can be an “impossible profession”
• The surprising literature on therapeutic effectiveness
• … and many, many things in between

For more conversations like this, be sure to subscribe to our Youtube channel (@ThinkingMachinesPodcast) and to our show in your favorite podcast player.

Artificial neural networks were designed to emulate the human brain - and their insane performance on a wide range of tasks is pretty good evidence to support the comparison.

Well, it's a bit more complicated than that, at least according to my guest Mark Ungless, former neuroscience lecturer at Imperial and Oxford and current Director of AI at the UK's Mental Health Innovations. Mark and I have collaborated on research for 5+ years, and I've long enjoyed his thoughts on the biology of the brain, the philosophy of the mind, and how AI is changing (and being changed) by our understanding of both. In our latest conversation, we touch on...

• His groundbreaking research on the biological root of loneliness
• How understanding neuroscience helps you understand humans
• How well neural networks represent our own minds (and whether it matters at all!)
• AI’s ability to do therapy
• What it means to have free will in a world where we understand more about our brains every day

For more conversations like this, be sure to follow our Youtube channel (@ThinkingMachinesPodcast) and subscribe to our show in your favorite podcast player.

Is the AI revolution we're experiencing going to push us into a future we can't imagine? Or will the pace of progress enable us to adjust along the way?

Dan Shipper spends his time thinking and writing on these topics (and many others) as the founder and CEO of Every Media, a technology-focused publication trying to understand the future. Dan is also a lifelong coder and entrepreneur with a background in philosophy and its intersection with tech. Dan and I share a ton in common (beyond just our first names!), so I think you all will enjoy our thoughts on...

• The meaning of the "good life"
• Whether the study of philosophy evolves over time
• How cryptocurrency, AI, and the nature of knowledge will accelerate each other
• The value of philosophy and thinking for thinking's sake
• What the future holds for engineering, philosophy, and humanity at large

If you like this conversation, check out more of Dan's thought at Every.to and on X 'at' @danshipper.

GPT-3 didn't have much of a splash outside of the AI community, but it foreshadowed the AI explosion to come. Is o1 OpenAI's second GPT-3 moment?

Machine Learning Researchers Guilherme Freire and Luka Smyth discuss OpenAI o1, it's impact, and it's potential. We discuss early impressions of o1, why inference-time compute and reinforcement learning matter in the LLM story, and the path from o1 to AI beginning to fulfill its potential.

00:00 Introduction and Welcome
00:22 Exploring O1: Initial Impressions
03:44 O1's Reception
06:42 Reasoning and Model Scaling
18:36 The Role of Agents
27:28 Impact on Prompting
28:43 Copilot or Autopilot?
32:17 Reinforcement Learning and Interaction
37:36 Can AI do your taxes yet?
43:37 Investment in AI vs. Crypto
46:56 Future Applications and Proactive AI

Dev Rishi is the founder and CEO of Predibase, the company behind Ludwig and LoRAX. Predibase just released LoRA Land, a technical report showing 310 models that can outcompete GPT-4 on specific tasks through fine-tuning. In this episode, Dev tries (pretty successfully) to convince me that fine-tuning is the future, while answering a bunch of interesting questions, like:

• Is fine-tuning hard?
• If LoRAX is a competitive advantage for you, why open-source it?
• Is model hosting becoming commoditized? If so, how can anyone compete?
• What are people actually fine-tuning language models for?
• How worried are you about OpenAI eating your lunch?

I had a ton of fun with Dev on this one. Also, check out Predibase’s newsletter called fine-tuned (great name!) and LoRA Land.

Talfan Evans is a research engineer at DeepMind, where he focuses on data curation and foundational research for pre-training LLMs and multimodal models like Gemini. I ask Talfan: 

• Will one model rule them all?
• What does "high quality data" actually mean in the context of LLM training?
• Is language model pre-training becoming commoditized?
• Are companies like Google and OpenAI keeping their AI secrets to themselves?
• Does the startup or open source community stand a chance next to the giants?

Also check out Talfan's latest paper at DeepMind, Bad Students Make Good Teachers.

"Understanding what's going on in a model is important to fine-tune it for specific tasks and to build trust."

Bhavna Gopal is a PhD candidate at Duke, research intern at Slingshot with experience at Apple, Amazon and Vellum.

We discuss

• How adversarial robustness research impacts the field of AI explainability.
• How do you evaluate a model's ability to generalize?
• What adversarial attacks should we be concerned about with LLMs?