brain – Earlybirds Invest

When you hear the word “neurotechnology,” you may picture Black Mirror headsets prying open the last private place we have — our own skulls — or the cyber-samurai of William Gibson’s Neuromancer. That dread is natural, but it can blind us to the real potential being realized in neurotech to address the long intractable medical challenges found in our brains. In just the past 18 months, brain tech has cleared three hurdles at once: smarter algorithms, shrunken hardware, and — most important — proof that people can feel the difference in their bodies and their moods.

A pacemaker for the brain

Keith Krehbiel has battled Parkinson’s disease for nearly a quarter-century. By 2020, as Nature recently reported, the tremors were winning — until neurosurgeons slipped Medtronic’s Percept device into his head. Unlike older deep-brain stimulators that carpet-bomb movement control regions in the brain with steady current, the Percept listens first. It hunts the beta-wave “bursts” in the brain that mark a Parkinson’s flare and then fires back millisecond by millisecond, an adaptive approach that mimics the way a cardiac pacemaker paces an arrhythmic heart.

In the ADAPT-PD study, patients like Krehbiel moved more smoothly, took fewer pills, and overwhelmingly preferred the adaptive mode to the regular one. Regulators on both sides of the Atlantic agreed: The system now has US and EU clearance.

Because the electrodes spark only when symptoms do, total energy use is reduced, increasing battery life and delaying the next skull-opening surgery. Better yet, because every Percept shipped since 2020 already has the sensing chip, the adaptive mode can be activated with a simple firmware push, the way you’d update your iPhone.

Scientists applied the same listen-then-zap logic farther down the spinal cord this year. In a Nature Medicine pilot, researchers in Pittsburgh laid two slender electrode strips over the sensory roots of the lumbar spine in three adults with spinal muscular atrophy. Gentle pulses “reawakened” half-dormant motor neurons: Every participant walked farther, tired less, and — astonishingly — one person strode from home to the lab without resting.

Half a world away, surgeons at Nankai University threaded a 50-micron-thick “stent-electrode” through a patient’s jugular vein, fanned it against the motor cortex, and paired it with a sleeve that twitched his arm muscles. No craniotomy, no ICU — just a quick catheter procedure that let a stroke survivor lift objects and move a cursor. High-tech rehab is inching toward outpatient care.

Mental-health care on your couch

The brain isn’t only wires and muscles; mood lives there, too. In March, the Food and Drug Administration tagged a visor-like headset from Pulvinar Neuro as a Breakthrough Device for major-depressive disorder. The unit drips alternating and direct currents while an onboard algorithm reads brain rhythms on the fly, and clinicians can tweak the recipe over the cloud. The technology offers a ray of hope for patients whose depression has resisted conventional treatments like drugs.

Thought cursors and synthetic voices

Cochlear implants for people with hearing loss once sounded like sci-fi; today more than 1 million people hear through them. That proof-of-scale has emboldened a new wave of brain-computer interfaces, including from Elon Musk’s startup Neuralink. The company’s first user, 30-year-old quadriplegic Noland Arbaugh, told Wired last year he now “multitasks constantly” with a thought-controlled cursor, clawing back some of the independence lost to a 2016 spinal-cord injury. Neuralink isn’t as far along as Musk often claims — Arbaugh’s device experienced some problems, with some threads detaching from the brain — but the promise is there.

On the speech front, new systems are decoding neural signals into text on a computer screen, or even synthesized voice. In 2023 researchers from Stanford and the University of California San Francisco installed brain implants in two women who had lost the ability to speak, and managing to hit decoding times of 62 and 78 words per minute, far faster than previous brain tech interfaces. That’s still much slower than the 160 words per minute of natural English speech, but more recent advances are getting closer to that rate.

Guardrails for gray matter

Yes, neurotech has a shadow. Brain signals could reveal a person’s mood, maybe even a voting preference. Europe’s new AI Act now treats “neuro-biometric categorization” — technologies that can classify individuals by biometric information, including brain data — as high-risk, demanding transparency and opt-outs, while the US BRAIN Initiative 2.0 is paying for open-source toolkits so anyone can pop the hood on the algorithms.

And remember the other risk: doing nothing. Refusing a proven therapy because it feels futuristic is a little like turning down antibiotics in 1925 because a drug that came from mold seemed weird.

Twentieth-century medicine tamed the chemistry of the body; 21st-century medicine is learning to tune the electrical symphony inside the skull. When it works, neurotech acts less like a hammer than a tuning fork — nudging each section back on pitch, then stepping aside so the music can play.

Real patients are walking farther, talking faster, and, in some cases, simply feeling like themselves again. The challenge now is to keep our fears proportional to the risks — and our imaginations wide enough to see the gains already in hand.

A version of this story originally appeared in the Good News newsletter. Sign up here!

]]>

The following is a guest post and opinion of Matej Janež, Head of Partnerships at Oasis.

At EthDenver earlier this year, one topic kept coming up again and again: AI and autonomous AI Agents. That excitement has carried over into other crypto conferences as the year has gone on. There is good reason for the excitement, too: these aren’t just ideas anymore – they’re here, and they’re handling real funds – but their reliance on transparent blockchains may become their biggest weakness.

What are these AI agents exactly? They’re smart software programs that work on their own to handle specific tasks. In crypto, they can use machine learning and blockchains to watch markets, spot patterns, and make trades automatically. Unlike traditional trading bots, today’s AI agents are adaptive; they refine their behavior continuously based on what yields results.

But there’s a big problem that has gone underappreciated and misunderstood: the fact that these onchain agents work on transparent blockchains makes their decision-making — their “brains” — essentially public. This openness creates real obstacles for agents trying to compete in financial markets.

AI Agents in DeFi

Right now, DeFi agents handle trading across decentralized exchanges, manage lending, and optimize yield farming. They react to market changes instantly, often making quick decisions with lots of money. Smart. Fast. Efficient.

But they face a basic challenge. The very system that lets them operate – public blockchains – shows their strategies to everyone. Every transaction, every interaction with a smart contract, leaves a trail that reveals how they “think”. It’s not much different from playing poker with your cards face up on the table.

Of course, one could run these strategies on private servers and only submit the final transactions to the blockchain, but this fundamentally defeats the purpose of crypto’s promise of transparency and onchain verifiability. The entire point of DeFi is to remove the need for trusted third parties and centralized systems.

Consider what’s already happening in DeFi today. A yield farming bot continually scans for the best returns across protocols, moving millions between lending platforms based on subtle market shifts. If its strategy becomes visible on-chain, competitors simply watch which pools it enters and exits, at what thresholds, and with what timing—then clone the strategy without the research costs. In decentralized credit markets, AI agents that score wallets for under-collateralized loans become pointless if borrowers can see exactly which behaviors improve their scores, leading to artificial wallet patterns designed to game the system.

Most concerning might be DAO treasury agents—when their rebalancing strategy is transparent, anyone can front-run major liquidity moves, effectively stealing from the community with each transaction. These aren’t edge cases; they’re fundamental flaws in applying AI to transparent systems where strategy execution and strategy development are impossible to separate.

Arguably, worst of all is the potential for market manipulation. When bad actors understand how an agent makes decisions, they can create situations designed to trick it. Markets full of transparent agents are easy targets.

Why a “Private Brain”?

A “private brain” for DeFi agents would fix these problems. By keeping computations confidential, agents could make decisions without showing their logic or intentions until transactions go through.

The security benefits are obvious. Strategies stay protected from copycats. Front-running becomes harder without seeing pending transactions. The agent’s work stays private. Teams that build better algorithms get to keep their edge, creating reasons to keep improving. The market rewards actual improvement instead of fast copying. On a larger scale, markets would become more stable. When agent strategies stay secret, you avoid herding – where multiple agents follow identical strategies. This cuts down on correlated market movements and lowers system-wide risk.

If we keep going like we are now – if DeFi agents keep operating with glass-box brains, we should be worried about a few things happening.

Market exploits will become more common and sophisticated. As agents handle more funds, the rewards for exploiting them grow too. Without privacy measures, these exploits become simple technical exercises rather than difficult security breaches.

Strategy cannibalization is just as worrying. When winning strategies get copied quickly, they stop working as well. Eventually, all agents use similar approaches, creating a monoculture. The market loses variety and resilience.

This leads to what you could all the “Hive Mind” problem; when all agents work the same way, they will react to market changes the same way too. This makes market swings bigger, increases volatility, and creates the risk of flash crashes when conditions trigger widespread identical responses. What starts as individual agents becomes, basically, one massive entity with system-wide effects. To spell it out: these are not the ingredients for a healthy market.

Technical Solutions

Trusted Execution Environments (TEEs) offer a solid way to create these private brains. TEEs provide secure areas where computation happens in isolation, protected even from the system hosting it. You can verify the work happened correctly, but the details stay private.

This tech lets us balance openness and privacy. The framework of an agent can be public and verifiable, while the specific decision-making and strategy details stay protected.

Adding private computation to DeFi agents isn’t just helpful—it’s necessary for algorithmic finance to grow properly. Without privacy, we’re building a market where innovation gets punished, exploitation gets rewarded, and system risks pile up under the surface.

We’re at a critical juncture in AI-powered finance where our choices will determine whether autonomous agents create a more efficient market or a dangerously fragile one. The technology for private computation exists today, but implementing it requires deliberate action from builders and protocols alike. As financial intelligence moves increasingly on-chain, ensuring these systems can operate with computational privacy won’t just protect individual strategies—it will safeguard the integrity of the entire DeFi ecosystem.