🕴🏻AI Candidate Shakes Up UK Elections

Ukraine's AI-powered drone assassination vision highlights urgency for binding regulations 

Key Points 

• At a recent NATO event, Ukraine's deputy tech minister discussed a prototype military drone equipped with advanced sensors and AI, enabling it to identify targets by voice and potentially carry out assassinations.

• The situation highlights that non-binding principles and ad hoc policies are insufficient in tackling the significant challenges posed by autonomous weapons.

☕News - Last week at a NATO event in Poland, Alex Bornyakov, Ukraine’s deputy tech minister, discussed a striking new development. He described a military drone equipped with advanced sensors and artificial intelligence capable of identifying a person by their voice. Once identified, the drone could execute a targeted assassination, such as taking out a Russian "war criminal."

This unsettling advancement in weaponized drones is currently in the “prototyping” phase within Ukraine. However, much of the necessary artificial intelligence needed for it already exists, Bornyakov stated.

This idea builds on Ukraine's innovative warfare strategies. The country has installed thousands of mobile phones on cell towers and gas stations to serve as digital ears. Data from these sensors is paired with a neural network to create AI tools that can track enemy drones or detect when Russia fires rockets.

🤖Ethical concerns surrounding Ukraine's advancements - It's worth noting that the idea of giving computers potential control over lethal decisions is controversial among Ukraine's allies.

In line with this, NATO has an ethical framework for AI, emphasizing the need for reasonable human input in any lethal use of force. There are even calls for these principles to be turned into legally binding rules, including by the United Nations.

The organization Stop Killer Robots further highlighted this concern, stating, “Non-binding principles and declarations, and ad hoc policy measures, are not sufficient to address the significant challenges which autonomous weapons pose.”

AI Steve: The surprise candidate shaking up UK's parliamentary election

Key Points 

• AI Steve, an avatar of real-life businessman Steven Endacott, is running as an Independent candidate for Parliament.

• What sets AI Steve apart is its interactive nature – voters have the opportunity not only to cast their ballots in its favor but also to engage in discussions about policies and raise concerns, all of which AI Steve incorporates into its platform.

• If it wins, Endacott will represent AI Steve in meetings and parliamentary sessions, where he will primarily facilitate discussions while leaving policy decisions to the AI. He envisions AI Steve as a means to enable more direct democracy, using AI as a supportive tool to better connect politicians with their constituents. 

🤵🏻‍♂️News - In the lead-up to the UK's upcoming elections, there's been an interesting addition to the candidate lineup: AI Steve, an avatar of real-life businessman Steven Endacott, that is running as an Independent candidate for Parliament.

What's unique about this candidacy is its interactive approach – voters have the opportunity not only to cast their ballots in AI Steve's favor but also to engage in discussions about policies and raise concerns, all of which AI Steve incorporates into its platform.

👨🏻‍💻For context - AI Steve was developed by Neural Voice, a company chaired by Endacott. According to Jeremy Smith, the company's co-founder, AI Steve can hold up to 10,000 conversations at once. This capability is crucial, as it allows AI Steve to engage with voters on a large scale, build a personalized database, and incorporate voter feedback into its platform.

AI Steve was born out of Endacott’s frustration with the political system when trying to advocate for important issues like the environment. Endacott believes significant government changes are needed to combat climate change and felt that to effect real policy change, one must be involved from the inside. In his previous attempts to run for office, he was disheartened by the emphasis on party politics and safe seats, rather than addressing the needs of the people.

Endacott claims AI Steve will stand out by transcribing and analyzing voter conversations, then presenting policy issues to "validators" – ordinary people who can express their support for specific issues or policies they want implemented.

⭐What's the goal? If it wins, Endacott will represent AI Steve in meetings and parliamentary sessions, where he will primarily facilitate discussions while leaving policy decisions to the AI. He envisions AI Steve as a means to enable more direct democracy, using AI as a supportive tool to better connect politicians with their constituents. 

To this end, Endacott's team plans to connect with commuters at the Brighton train station, situated roughly an hour outside of London. They will ask commuters to participate in brief policy surveys via email during their journeys to or from the city.

Researchers test reliability of AI in medicine... the result: concerning!

Key Points 

• Researchers have warned that LLMs and LMMs cannot be trusted when it comes to high-stakes, real-world medical scenarios, at least not yet.

• To gauge the reliability of AI models, the researchers assessed them on a range of diagnostic questions.

• They found an average drop of 42% in model performance. 

🏥Context of the news - Large language models (LLMs) and large multimodal models (LMMs) are increasingly being integrated into medical settings. However, these groundbreaking technologies have yet to undergo thorough testing in such critical areas, raising concerns about how reliable they really are. 

👨🏻‍🔬News - Researchers from the University of California, Santa Cruz, and Carnegie Mellon University have now warned that LLMs and LMMs cannot be trusted when it comes to high-stakes, real-world scenarios, at least not yet. 

🤕Here's why - In their recent study, the researchers set out to see how dependable these models are when it comes to medical diagnosis. They did this by asking a range of general and specific diagnostic questions.

They gathered a fresh set of data and put state-of-the-art models to the test with X-rays, MRIs, and CT scans of different body parts like the abdomen, brain, spine, and chest. What they found was concerning: 

• Advanced models like GPT-4V and Gemini Pro exhibited performance comparable to random educated guesses when attempting to recognize conditions and positions.

• Furthermore, the introduction of adversarial pairs, which are slight perturbations, notably decreased model accuracy.

• On average, the accuracy of the tested models decreased by 42%.

Here's what Xin Eric Wang, a professor at UCSC and paper co-author, posted to X: