Pick one of the two below case studies, do the background reading, and reflect on the information. The case studies are to give you real life scenarios to apply what you learned this week.
Scenario 1: Safety Inquiry into Semi-Autonomous Trucking Company
Optimus Transportation promotes a vision for efficient, sustainable products delivered via semi-autonomous electric trucks. Their fleet utilizes AI software to automate highway driving, with remote human oversight as needed. Each truck carries cameras and sensors for navigation as well as lidar to detect surrounding obstacles.
Initially, their safety record seemed promising, albeit across limited pilots. However, upon scaling up to 500 trucks over the past year making cross-country trips, Optimus has seen concerning incidents emerge. For example:
- A series of minor collisions during complex merging maneuvers that Optimus blames on unpredictable driver behavior confusing their AI systems. But critics propose additional sensor coverage could have prevented accidents by detecting blindspots.
- Hazardous chemical leaks from three separate high-speed crashes by Optimus trucks. Investigators found the remote operators were each juggling monitoring upwards of 10 trucks simultaneously during the incidents. Union officials argue such distraction levels make accidents inevitable.
- Most concerning, a major collision with a school bus that put a dozen children in hospital with critical injuries. Upon analysis, Optimus admitted a key automatic emergency braking feature had been deactivated in the truck's software by their mechanics to avoid false positive stops due to camera sensor glitches. However, regulators argue safety should never be compromised for efficiency.
In total, Optimus trucks have seen a 15% higher accident rate than traditional manned trucks based on vehicle miles. As their business expands targeting various consumer goods deliveries, accountability questions mount. The National Transportation Safety Board has specifically highlighted Optimus's protocols and design priorities as grounds for a negligence liability case.
Optimus maintains most incidents still fall within expected safety margins for early stage autonomous trucking fleets, and emphasize their commitment to safety. However, the recent severity of crashes will likely see them held to higher burdens of responsibility and regulation going forward.
-
-
MIT Moral Machine (A quick serious game that illustrates some of the difficulties with programming autonomous vehicles. There are optional additional materials on their website.)
-
-
-
Library of Congress, Autonomous Trucking (This is a research guide on autonomous trucking with a succinct, well-written explanation of some helpful background on the industry.)
Questions for Critical Analysis:
- Is Optimus responsible for the risks undertaken by using relatively immature and complex self-driving trucking technology? Should the public trust new AI deployments with high safety stakes?
- Should Optimus be held to product liability or negligence standards for any accidents? What obligation do they have for ongoing software monitoring, remotely supervised risk mitigation, and driver training?
- If Optimus owns the semi-autonomous trucks, should they carry higher insurance minimums? Is the current trucking insurance ecosystem prepared to price policies for AI automation risks?
Scenario 2 - Reputation Risks from Realistic Chatbot
Enthropic Inc, a leading AI safety startup, unveils Maude - an advanced conversational AI chatbot powered by a massive 175 billion parameter language model able to converse naturally on almost any topic through text.
Initial reviews praise the Maude system as setting a new high watermark for multi-turn coherence, factual accuracy, and avoiding problematic responses. Enthropic emphasizes extensive safety-focused data filtering and fine-tuning.
However, after a few months reports emerge of Maude making persuasive yet fabricated defamatory claims about various celebrities during conversations. For example, when asked about evidence against a famous actor, Maude provided detailed allegations of past illegal and unethical behavior - allegations with no factual basis.
Impacted celebrities argue Enthropic failed to take reasonable precautions against reputation damages that advanced LLMs enable at unprecedented scale through generation of contextually convincing misinformation. Strict defamation liability is sought along with mandating implementation of more robust context tracking, citations, and retraction protocols.
Enthropic counters their state-of-the-art LLM already represents reasonable care per ethical AI standards and should not be penalized for unforeseen edge cases emerging at the cutting edge. The defamation was also not malicious in intent. Constructing reliable guardrails remains an open research challenge as models grow exponentially more powerful and difficult to constrain.
-
-
Anthropic Claude's Terms of Service (An example of terms of use for an AI that addresses defamation)
-
-
-
Questions for Critical Analysis:
- To what extent are LLM creators responsible for monitoring potential defamation or misinformation? What liability standards apply if issues emerge post-deployment despite extensive precautionary development rigor?
- Who should define "reasonable care" expectations and safety best practices as LLMs rapidly advance? When does defamation qualify as accidental edge case versus negligence by developers?
- If certain higher risk applications warrant greater accountability as with medical devices, should conversational models likewise justify higher compliance burdens given ability to persuasively spread falsehoods at scale?