🧪Building AI with Biology

Scientists turn to biocomputing in bid to create energy-efficient AI

Key Points 

• Scientists are using lab-grown cells, called organoids, to develop AI systems that are more energy-efficient and address data storage needs. 

• Swiss tech company FinalSpark’s 'Neuroplatform' is a computer platform powered by human brain organoids.

• Key limitations include the lack of standardized manufacturing and the limited lifespan of organoids, which lasts about 100 days.

👨🏻‍🔬News - In an effort to create more energy-efficient AI, some scientists are turning to biocomputing, a field that involves using living organisms to build computers. Both academic and commercial researchers are exploring this approach to address the growing energy and data storage demands of AI. 

ℹ️Biocomputing explained - Simply put, this approach involves using synthetic biology, like tiny clusters of lab-grown cells known as organoids, to build computer architecture.

😎A case in point - Swiss tech company FinalSpark recently introduced "Neuroplatform," a computer platform powered by human brain organoids. Scientists can rent access to it online for $500 a month.

Neuroplatform operates with processing units that each contain four spherical brain organoids, each measuring 0.5 millimeters in diameter. These organoids are connected to eight electrodes, which both electrically stimulate the neurons and link the organoids to conventional computer networks. The neurons are also exposed to dopamine to simulate the brain's natural reward system. 

This combination of dopamine and electrical stimulation helps train the organoids, encouraging them to form new neural pathways, similar to how the human brain learns. FinalSpark suggests that with further development, this training could enable organoids to mimic the functions of silicon-based AI and eventually serve as processing units similar to today's CPUs and GPUs.

🤕Limitations - One major limitation is the lack of a standardized manufacturing process. Additionally, since living brains eventually die, FinalSpark's organoids only survive for about 100 days. This means an AI system could be trained and run for a few months before needing to be transferred to a new cluster or risk being lost permanently.

Researchers propose "personhood credentials" to combat AI bots

Key Points 

• Researchers suggest in-person verification to issue “personhood credentials,” potentially limiting AI bot activity online.

• The system aims to prove human identity without revealing personal details, though security against cyberattacks is uncertain.

🤖News - In a recently published paper, researchers expressed concerns about AI bots potentially taking over the internet and spreading like a digital invasive species. Rather than attempting to limit these bots and AI-generated content, the paper proposed a unique solution: requiring individuals to be verified in person by another human to obtain “personhood credentials.”

The main idea is to develop a system that allows people to prove they are human without disclosing their identity or personal details. This concept is similar to “proof of personhood” technologies found in the blockchain community.

Researchers are proposing a system where specific organizations or facilities act as issuers to verify an individual's humanity. These issuers would use human staff to confirm and certify each person's credentials, without tracking how the credentials are used. However, the system’s resilience against cyberattacks and quantum-assisted decryption remains uncertain.

On the flip side, service providers could choose to grant accounts only to those with verified credentials, effectively limiting each person to one account and preventing bots from accessing these services.

🤔Why this matters - As of August 2024, ChatGPT’s safeguards would likely stop it from being used to create multiple free Reddit accounts. While some AI can bypass CAPTCHA checks, navigating the full email verification and account setup process would be complex. The key argument from the team, which included experts from OpenAI, Microsoft, a16z Crypto, and institutions like Harvard, Oxford, and MIT, was that current protections might only be effective for a limited period.