3 ms·

There are a few broad areas of risk in AI. 1. Enabling goes both ways, therefore bad actors can also be enabled by AI. 2. Accuracy of suggestions. Informati

by Lerc 2y ago

There are a few broad areas of risk in AI. 1. Enabling goes both ways, therefore bad actors can also be enabled by AI. 2. Accuracy of suggestions. Information provided by AI may be incorrect, be it code, how to brush one's teeth, or height of Arnold Schwarzenegger. At worst AI can respond against the users interests if the creator of the AI has configured it to do so. 3. Accuracy of Determinations. LLM-as-a-Judge falls under this criteria. This is one of the areas where a single error can magnify the most. This post says: What about guardrails? Guardrails are a separate but related topic. They are a way to prevent the LLM from saying/doing something harmful or inappropriate. This blog post focuses on helping you create a judge that’s aligned with business goals, especially when starting out. That seems woefully inadequate. When using AI to make determinations there has to be guardrails. Having looked at drafts of legislation and position statements of governments, many are looking at legally requiring that any implementers of AI systems that make determinations must implement processes to deal with the situation where the AI makes an incorrect determination. To be effective this should be a process that can be initiated by individuals affected by this determination.

nine_zeros 2y ago
> Having looked at drafts of legislation and position statements of governments, many are looking at legally requiring that any implementers of AI systems that make determinations must implement processes to deal with the situation where the AI makes an incorrect determination The real legislation we need is liability. Who is liable to suffering caused by LLMs inaccuracies? I think if liability should be on corporations selling the LLM as a solution. If a person gets arrested for selling the police a fuzzy LLM solution, and this causes unnecessary grief to the individual, the seller of the LLM service must compensate the individual with 4x the median income of their metropolitan area, for the duration of the harm caused.
- phs318u 2y ago
  > liability should be on corporations selling the LLM as a solution I agree. There was a period of time (don't know if it's still the case), where Microsoft Windows licenses/ToS expressly and explicitly did not warrant the use of the OS for a whole bunch of critical use-cases (e.g. anything with the word 'nuclear' in it). If the vendor doesn't warrant it for that use, then you'd be remiss in your duty to your company to ignore that and choose to use it for that purpose. Caveat emptor.
salawat 2y ago
>1. Enabling goes both ways, therefore bad actors can also be enabled by AI. >2. Accuracy of suggestions. Information provided by AI may be incorrect, be it code, how to brush one's teeth, or height of Arnold Schwarzenegger. At worst AI can respond against the users interests if the creator of the AI has configured it to do so. >3. Accuracy of Determinations. LLM-as-a-Judge falls under this criteria. This is one of the areas where a single error can magnify the most. sed s_AI_technology_g < quoted_text
- Lerc 2y ago
  That only really applies to 1. Technology in general does not have the ability to advise. Advice is already a legally protected category in many fields. Legal, medical, and financial advice are each covered by their own legislation to require the advisers to act in the interest of their client. Technology in general has been used for years to make determinations based upon physical characteristics. Weighs more than 50g, Temperature over 20C, Item A is heavier than Item B. The distinction with AI is that it will be making semantic determinations. Computers have also been used to implement policy decisions in a manner that makes them appear to be making determinations. When someone says "Computer says no", it's usually a human who made a decision. Attributing it to the computer just makes it impossible to argue against. The world would probably benefit from there being regulation on processes to enable people to challenge incorrect determination in this area as well. We have all seen the news stories of a big tech company cutting a user from a vital service because of an arbitrary seeming decision. Often the only challenge mechanism that is effective is when the situation receives widespread publicity. While the decision causing the issue is usually human in these instances, it is shielded behind a wall of technology leaving the user with no recourse.

by Lerc 2y ago

nine_zeros 2y ago

> Having looked at drafts of legislation and position statements of governments, many are looking at legally requiring that any implementers of AI systems that make determinations must implement processes to deal with the situation where the AI makes an incorrect determination The real legislation we need is liability. Who is liable to suffering caused by LLMs inaccuracies? I think if liability should be on corporations selling the LLM as a solution. If a person gets arrested for selling the police a fuzzy LLM solution, and this causes unnecessary grief to the individual, the seller of the LLM service must compensate the individual with 4x the median income of their metropolitan area, for the duration of the harm caused.

phs318u 2y ago
> liability should be on corporations selling the LLM as a solution I agree. There was a period of time (don't know if it's still the case), where Microsoft Windows licenses/ToS expressly and explicitly did not warrant the use of the OS for a whole bunch of critical use-cases (e.g. anything with the word 'nuclear' in it). If the vendor doesn't warrant it for that use, then you'd be remiss in your duty to your company to ignore that and choose to use it for that purpose. Caveat emptor.

salawat 2y ago

>1. Enabling goes both ways, therefore bad actors can also be enabled by AI. >2. Accuracy of suggestions. Information provided by AI may be incorrect, be it code, how to brush one's teeth, or height of Arnold Schwarzenegger. At worst AI can respond against the users interests if the creator of the AI has configured it to do so. >3. Accuracy of Determinations. LLM-as-a-Judge falls under this criteria. This is one of the areas where a single error can magnify the most. sed s_AI_technology_g < quoted_text

Lerc 2y ago
That only really applies to 1. Technology in general does not have the ability to advise. Advice is already a legally protected category in many fields. Legal, medical, and financial advice are each covered by their own legislation to require the advisers to act in the interest of their client. Technology in general has been used for years to make determinations based upon physical characteristics. Weighs more than 50g, Temperature over 20C, Item A is heavier than Item B. The distinction with AI is that it will be making semantic determinations. Computers have also been used to implement policy decisions in a manner that makes them appear to be making determinations. When someone says "Computer says no", it's usually a human who made a decision. Attributing it to the computer just makes it impossible to argue against. The world would probably benefit from there being regulation on processes to enable people to challenge incorrect determination in this area as well. We have all seen the news stories of a big tech company cutting a user from a vital service because of an arbitrary seeming decision. Often the only challenge mechanism that is effective is when the situation receives widespread publicity. While the decision causing the issue is usually human in these instances, it is shielded behind a wall of technology leaving the user with no recourse.