Be Careful what you ask for...

Bruce Schneier cautions us that Large Language Models LLMs are very literal in their interpretation of what they are asked to do.  They are very good at interpreting the words and very poor at divining or understanding the intent or meaning.  Midas wanted gold but got the Golden Touch.  Sissa ben Dahir wanted a big reward for inventing chess but he really did not intend to destroy the world.  

One problem in large language models (LLMs) is the use of human language to specify the goals.  Bruce, and his colleague, Barath Raghavan, speak of the genie coefficient, the gap between what a user intends and how the AI interprets it.  The proper interpretation of human language relies upon context.  "If you ask  someone to 'get coffee,' they bring you a cup, not a plantation."  

Human language is not a good medium for fully specifying a goal or a result. We are not good at using it for that and machines are not good at interpreting it.  Because a prompt lacks any context, or even history, the prompt must include any necessary context.  That is why using Claude to write code is an iterative process of refining the results "until the user is satisfied." It is why IBM prefers application specific AIs that may use language for communicating with humans but not for communicating with each other.  It is why giving the same prompt to two different AIs, or even to the same one on different occasions, may render different results.  

We cannot allow ourselves the error of believing that because it can mimic the way that we communicate that it thinks like us.  Perhaps we should reinterpret the first "L" in LLM to stand for "limited" rather than "large."  

Responsibility and Accountability for AI

A human being or enterprise is, and must be held, responsible for everything that a computer is asked to do.  For AI they are responsible for the training, goals, data, weights, and, prompts.  They are responsible for all the results, i.e. action, uses, and properties including accuracy, correctness, and purpose.

The responsibility must include criminal or civil accountability.  Consider Anthropic's attack against Hugging Face. Anthropic lacked Mens Rea so they were not criminally liable.  However, they were reckless and therefore they are civilly liable for any and all damage that their experiment caused.  It is essential that they not be allowed to escape that responsibility.  We must limit criminal and reckless use of what it only a tool.  

We worry about AI going rogue but that is not the problem.  The problem is what someone or some enterprise asked the AI to do and what precautions they took.  It is clear that we are not very good at that.  If we want to enjoy the benefits of AI while avoiding unintended consequences we must hold each other scrupulously accountable for its uses.  

We must restrict the use of LLMs to applications where we can predict, evaluate, and control the results. I continue to recommend IBM's stated approach: application specific uses, curated, application, training data, and transparency and governance built in from the ground up.  I like the approaches of Apple Siri and Microsoft Co-pilot, that use AI to assist with specific tasks in a known context.   

What am I missing?

Artificial Intelligence

Artificial intelligence, AI, is a new user interface to the computer.  Large language models (LLMs) make the computer easier to use.  They permit us to describe the result that we want in natural language. improving productivity and enabling new applications.  AI will improve intellectual productivity as much as the internal combustion did for manual productivity.  In response to internal combustion, and more specifically the tractor,  we shortened the work week from 72 hours to 44 and invented vacations and retirement.  In the process, we killed off two generations of young men and still suffered 25% unemployment.  Said another way, increases in productivity are disruptive.  

The computer, with or without AI, is a tool.  Tools vary in quality, utility, usability, and use.  The user is responsible for the selection of the tool, the purpose to which it is put, and for all the properties of the result.  This is true whether the user is an individual or a group.  An enterprise must be responsible and accountable for everything that results from its application of this powerful technology. We call this security and we forget any part of it at our peril. We must not impute authority or autonomy to the tool; we must not anthropomorphize the tool.  "The craftsman does not blame his tools."  We must hold people accountable for how we use this powerful new tool.  

In the near term we should focus on embedded application specific implementations of AI.  We should follow the example of IBM, a pioneer in the field.  IBM trains the engine, think Watson, on application specific curated data; they build in governance and transparency from the ground up.

Public policy must soften the impact of the disruption.  This will include shortening the work week to spread the work and the leisure.  It should include a guaranteed minimum income to ease transition from old jobs and skills to new ones.  Finally, it should include changes in tax policy from labor to capital, people to robots, and production to consumption, to more securely and equitably finance the social safety net.  

"Women and Children First"

As we approach autonomous cars, some have posed the question "In accident avoidance should the car prefer its passenger or the pedestrian?"  It is posed as a difficult ethical dilemma.  I even heard an engineer suggest that not only does he not want to make the decision but that he would like Congress to make it as a matter of law.  

This is really just another instance of an ethical dilemma that humanity has faced forever.  It has many illustrations but one that has been used in teaching ethics is called the "life boat" problem.  If there is not enough room in the lifeboat for everyone, who gets in?  If there is not enough food and water, who gets preference?

The simple answer is "women and children first."  Human drivers will steer away from a woman with a baby carriage even if they do not have time to evaluate the alternative.  It is built into the culture, all but automatic, but the reason is that it is pro life.  Children are the future of the species.  Women can nurture and reproduce.  Men can sow but they cannot reap.  While the male role takes minutes, the female role takes months.  Life needs more females than males.

The reason that we do not apply this pro life rationale to the autonomous automobile is that we assume that the consideration is beyond its capability.  However, most of what one expects of an autonomous car today was beyond its capability a decade ago.  For the moment, most may not be able to consider all the factors we might like.  For example, they may not recognize age and gender, much less consider them.  Ten years from now, they certainly will.  

In this context it is useful to consider how such systems make a decision.  They identify a number of scenarios, depending upon the time available, assign an outcome and a confidence level to each, and choose statistically.  The kind of ties implied by the strawman dilemma will be vanishingly rare, even more so as the computers become faster and the number of things the can consider increases.  

Compare the autonomous car to the human driver.  In the two tenths of a second that it takes a young adult to recognize and react, the autonomous car will evaluate dozens of possibilities with as many considerations.  Like the human driver, the autonomous car may confront instances when there are simply no good options but the whole reason for using them is that they are less likely than the human driver to overlook the least damaging.