Latest news about Bitcoin and all cryptocurrencies. Your daily crypto news habit.
![](https://cdn-images-1.medium.com/max/1024/1*8IW0UhSndjBCJ1kv2MpJ_w.jpeg)
As companies around the world race to develop AI-powered products, the technology’s role in daily life is becoming increasingly prominent: it can manage finances, help doctors diagnose diseases, and drive cars. However, AI models are effectively black boxes to most of us. We’ve no option but to blindly trust the underlying algorithms because we don’t have visibility into how they work.
To build trust, users must have the ability to verify how a model has been trained and the inference process that generates an output. ICP’s computational power and smart contract expressiveness are uniquely positioned to deliver this functionality.
The black box problem
Stanford University’s Artificial Intelligence Index Report 2024 found that nearly half of respondents are nervous about the use of AI for nefarious purposes. After gaining access through a backdoor vulnerability, hackers can tamper with the algorithm so it generates malicious outputs that serve their interests and mislead users. AI startup Anthropic recently published a paper demonstrating the severity of this risk using the example of ‘sleeper agents’. Anthropic programmed three large language models to behave maliciously under certain circumstances, which went unnoticed when the models launched but could be activated by a specific prompt (hence the espionage-inspired name). Any prompt entered in 2023 produced an accurate output, but when the year ticked over to 2024, the sleeper agent kicked in, and the model delivered erroneous results.
In March 2024, security firm Oligo detected an ongoing cyberattack targeting Ray, an open source framework used by thousands of developers, including OpenAI (creator of ChatGPT) and Amazon, to scale AI applications. The exploit, which started seven months previously, exposed details of AI production workloads which could allow the hackers to tamper with models during the training stage and access sensitive private data including credentials for accounts on OpenAI, Stripe, and Slack. The hackers also hijacked massive amounts of computing power from hundreds of companies which they used to mine cryptocurrencies.
The key takeaway is that malicious parties can tamper with AI models without alerting users and developers. Traditional techniques for assessing the integrity of software, such as source code analysis, are infeasible for AI models due to their size, so the industry needs to take an alternative approach to building trust.
AI on the blockchain
Decentralized AI, or DeAI for short, is the moniker used to describe the intersection between AI and blockchain technology. Some projects use the term for peripheral elements such as tokenization or decentralized marketplaces, but the truest form of DeAI runs models fully on-chain and leverages smart contracts. Here’s how ICP makes this possible:
- Security: Computation is replicated across multiple nodes and validated by ICP’s consensus mechanism which harnesses chain-key cryptography, a suite of advanced cryptographic mechanisms, to make AI models tamper-proof. There’s no single point of failure for hackers to target.
- Verifiability: Open source smart contracts allow users to verify how the models they contain use data for inference. The same level of transparency will apply to the training phase once smart contracts support it.
- Resilience: Smart contracts are always available and censorship resistant as they aren’t controlled by a single entity or legislation. The control structure can be decentralized in the form of a DAO or non-existent, in which case the smart contract doesn’t belong to anyone and its code is immutable.
DeAI is too compute and memory-intensive for traditional blockchain networks. But the advanced design of ICP, which combines security, scalability, and computation power means developers can currently run inference entirely on-chain. The long-term goal is to eliminate the black box problem once and for all by supporting the training of AI models in smart contracts using GPU-enabled nodes. Developers can also integrate open source projects, like the Sonos Tract AI inference engine, into DeAI models because WebAssembly, ICP’s smart contract runtime environment, supports a growing library of languages and tools.
To see DeAI on ICP in action, check out this demo presented by Dfinity’s Founder, President and Chief Scientist Dominic Williams, showing an AI model running as a smart contract (a world first) and correctly identifying a variety of images. And keep an eye out for the second part of this series, which will explore some potential DeAI use cases enabled by ICP.
Read more about DeAI on ICP : https://internetcomputer.org/ai
Disclaimer
The views and opinions expressed in this article are solely those of the authors and do not reflect the views of Bitcoin Insider. Every investment and trading move involves risk - this is especially true for cryptocurrencies given their volatility. We strongly advise our readers to conduct their own research when making a decision.