Why AI Can't Reason Over Time: A Bioengineer on Building What Doesn't Exist Yet

A career-ending soccer injury sent Michael Iwashima toward engineering. Years later, he's building brain-computer interfaces that let people with paralysis play Space Invaders, 3D-printable biosensors that help farmers detect crop disease, and a mold-detection startup aimed at a problem most homeowners can't see.

In this Still Human conversation, Michael talks through the limits he keeps running into with current AI. It's powerful at one-shot tasks, he says, but it can't reason over time. It doesn't reflect, doesn't remember, doesn't adjust the way humans do across days and weeks. That gap shapes how he builds.

Why AI Can't Reason Over Time

Modern AI is impressive at one-shot tasks — give it a prompt, get an answer. What it can't do is the thing humans do without thinking about it: carry context across days and weeks, notice a pattern that took months to emerge, change its mind based on something it learned yesterday. Michael keeps running into this limit when he tries to apply AI to longitudinal problems like accessibility and crop disease, and it's the limit that defines where he chooses to build.

Brain-Computer Interfaces & Accessibility

Some of the most meaningful AI work happens where the technology meets a body that needs it. Michael's BCI research translates neural activity into control signals so people with paralysis or cerebral palsy can interact with devices — including games like Space Invaders — without needing fine motor control. The framing isn't "what can AI do" but "where can AI actually expand what a human body can do."

Frugal Science in Agriculture

Most biosensors require expensive lab infrastructure. Michael's work — associated with the Verma Lab at Purdue — is on the opposite end: 3D-printable, low-cost sensors farmers can deploy in the field to detect crop disease. It's a category called frugal science, and it matters because the people who most need real-time biological data are the ones least likely to have access to a lab.

A Mold Detection Startup

Most homeowners only learn about mold after it's already a problem. Michael is building a startup to detect it before that point — a problem that sits in the same family as the BCI and agricultural work: AI applied where humans physically can't see what's happening. It's the kind of problem that makes the "useful vs flashy" question very easy to answer.

A "1 of 1" Education Model

If AI can transmit knowledge well, then the human role in education should shift to what AI can't do — curiosity, judgment, the parts of learning that don't compress into a tutorial. Michael's "1 of 1" framing imagines each student getting an AI knowledge layer plus a human mentor focused on the irreducibly human parts. Not a replacement of teachers; a redirection of what teachers spend their time on.

Community Building as a Frontier Skill

Michael helped grow AI Collaborate at Santa Clara from near-shutdown to 200 members in its second year. His read: in an AI-saturated world, access to information is no longer the bottleneck. The bottleneck is the people you build alongside — the room where the next idea actually gets shipped. That makes community building one of the most underrated capabilities of the next decade.

Show Notes

Michael Iwashima is a bioengineer and student founder working at the intersection of brain-computer interfaces, agricultural biosensors, and accessibility technology. He came to engineering after a soccer injury closed an earlier path. Today his work spans BCI research aimed at people with paralysis and cerebral palsy, frugal-science biosensors developed in association with the Verma Lab at Purdue, and a startup focused on mold detection. He also helped build AI Collaborate at Santa Clara University into a 200-member community after the club nearly folded in its first year. For the Still Human audience, Michael is the guest who pushes the AI conversation past hype and toward where the technology actually meets the human body.

Articles & Research

• Verma Lab (Purdue University) — Research group associated with Michael's work on 3D-printable, low-cost biosensors for agricultural applications

Tools & Resources

Relevant to this episode:

• Brain-computer interfaces (BCIs) — Hardware and software systems that translate neural activity into control signals, enabling people with paralysis to interact with devices and games like Space Invaders
• 3D-printable biosensors / frugal science — Low-cost, field-deployable sensors designed to detect crop disease and other biological signals without expensive lab infrastructure
• AI Collaborate (Santa Clara University) — Student community Michael helped grow from near-shutdown to ~200 members, focused on building and shipping AI projects together

Related Still Human Episodes

You might also enjoy:

• Execution Culture — Sean Wu on Raising $2M for Robotics & the Sim-to-Real Gap — oshenstudio.com/episode/execution-culture-sean-wu-synphony-robotics
• Build Before You're Ready — Andrey Marey on High-Agency, Discipline & Refusing AI With Friends — oshenstudio.com/episode/high-agency-andrey-marey-student-founder

People Mentioned

No additional people were cited by name in this episode beyond the host and guest.

Key Takeaways

• AI still can't reason over time. It's strong on one-shot tasks but doesn't reflect, remember, or adjust across days and weeks the way humans do. That gap is where Michael focuses his work.
• Execution culture beats ideas. AI Collaborate almost folded in its first year. What turned it around wasn't a better idea — it was follow-through.
• A "1 of 1" education model. Let AI handle knowledge transmission so humans can focus on curiosity, judgment, and the parts of learning that don't compress into a tutorial.
• Community building is underrated. In an AI-saturated world, the bottleneck isn't access to information; it's the people you build alongside.
• Accessibility teaches you what AI is for. Working on tech for people with paralysis and cerebral palsy reframes the question from "what can AI do" to "where does it actually expand what a human body can do."

In This Episode

• Why "AI can't reason over time" is the real limit — Michael's working frame for where the technology stops and where humans still matter
• Brain-computer interfaces for accessibility — BCI research aimed at people with paralysis, including controlling games like Space Invaders through neural signals
• Frugal science and AI in agriculture — 3D-printable, low-cost biosensors developed in association with the Verma Lab at Purdue, designed to detect crop disease in the field
• A mold-detection startup — Building a company around a problem most homeowners never see until it's too late
• From near-shutdown to 200 members — How AI Collaborate at Santa Clara University became a community after almost folding, and what that says about execution culture
• A "1 of 1" education model — Letting AI handle knowledge transmission so humans can focus on curiosity, judgment, and creativity
• Community building as a frontier skill — Why Michael thinks this is the most underrated capability in an AI-saturated world

About Michael Iwashima

Michael Iwashima is a bioengineer and student founder building at the intersection of brain-computer interfaces, agricultural biosensors, and accessibility technology. His path into engineering ran through a career-ending soccer injury, which gave him an early relationship with rebuilding from nothing. His current work spans BCI research aimed at people with paralysis and cerebral palsy, frugal-science biosensors developed in association with the Verma Lab at Purdue University, and a startup focused on mold detection. He also helped build AI Collaborate at Santa Clara University into a 200-member community after it nearly folded in its first year. For the Still Human audience, Michael is the guest who moves the AI conversation past hype and toward where the technology actually meets the human body — and where it still falls short.

Connect With Michael Iwashima

• Watch the episode: youtube.com/watch?v=GkyO4MgQW1k
• Add Michael's Instagram, LinkedIn, or personal website here

Follow Still Human Podcast

Still Human Podcast is a biweekly show by Oshen Studio, hosted by Perkin — exploring what it means to stay human in the age of AI. Real conversations with builders, creators, founders, and thinkers doing it in real life.

• YouTube: youtube.com/@Oshen.studio
• Instagram: instagram.com/perkin0909
• Listen everywhere: Search Still Human Podcast on Spotify, Apple Podcasts, or wherever you listen

New episodes drop every two weeks. Subscribe so you never miss a conversation.