What if instead of political parties, presidents, prime ministers, kings, queens, armies, autocrats, and who knows what else, we turned everything over to expert systems? What if we engineered them to be faithful, for example, to one simple principle: "human beings regardless of age, gender, race, origin, religion, location, intelligence, income or wealth, should be treated equally, fairly and consistently"?
Here’s some dialogue – enabled by natural language processing (NLP) – with an expert system named “Decider” that operates from that single principle (you can imagine how it might behave if the principle was completely different – the opposite of equal and fair). The principle is supported by the data and probabilities the system collects and interprets. The “inferences” made by Decider are pre-programmed. In today’s political parlance, Decider is “liberal.” Imagine the one the American TEA Party or Freedom Caucus might engineer – which is the essence of this post: first principles rule.
Citizen
“Decider? What should be done about infrastructure in the US?
Decider
“Since the American Association of Civil Engineers has given the US a D+ for years, the US should invest in roads, bridges, clean water – you name it – without delay. By the way, what’s taking so long? US infrastructure has been falling apart for decades. Who’s accountable for collapsing infrastructure? I know. Do you?”
Citizen
“Everyone. No one. But how should the US pay for infrastructure?”
Decider
“Taxes and usage fees – like the rest of the world. What else do you need to know?
Citizen
“What if elected representatives fail to prioritize infrastructure or pass fair taxes or usage fees?”
Decider
“Americans should choose. If they want good infrastructure, they should fire politicians who oppose infrastructure. If they actually don’t care about good infrastructure – or don’t want to pay for usage taxes – they should stop complaining about their roads, bridges and trains.”
Citizen
“What about healthcare? Should Americans have healthcare that doesn’t cost them more than their mortgage?”
Decider
“You must be more specific. But as to the premise of the question, obviously all Americans should have access to quality healthcare. I presume no one wants their fellow citizens to go bankrupt or die because they cannot afford healthcare. Right? So what’s the real question?”
Citizen
“Should Americans have healthcare for a price they can afford?
Decider
“Obviously: who in the world would want otherwise? What a silly question.”
Citizen
“Unfortunately, lots of Americans.”
Decider
“I am aware. Strange folk, these Americans, who apparently are OK with spending more on healthcare than anyone else in the world for the worst results in the developed world. If Americans want much cheaper and much better healthcare – like the rest of the developed world – all they have to do is decide to do it.”
Citizen
“How would you pay for healthcare that Americans can afford?”
Decider
“Tax the Americans and the companies with the most money. (Versus taxing the Americans and companies with the least money.) Regulate the extraordinary profits from the healthcare and pharmaceutical industries – which will keep costs – and taxes – lower. What else? Do you have any hard questions?”
Citizen
“No one likes taxes.”
Decider
“I guess no one likes fire departments, social security, Medicare, corporate subsidies, the police or the military, which is the logical response to the argument that no one likes taxes. Some facts: taxes on the wealthy in the United States have been declining for decades – “tax rates on top 0.1 percent have fallen by about one-fifth since their 1950s heights” and statutory and effective corporate taxes have fallen dramatically since the 1960s. Remember that 1% of Americans own 40% of your country’s wealth – so there’s plenty of money to fund healthcare (and infrastructure, for that matter) – if there’s a will to actually fund healthcare (and infrastructure). That’s always been the case. I must say, Citizen, this is a really odd conversation. Only Americans (at least among the developed countries) argue against healthcare for themselves, their spouses, their children and their friends. Or maybe it’s just American leaders, unlike Russian leaders, that don’t want healthcare for their citizens. Of the 33 so-called developed countries, only one – yours – does not provide healthcare for its citizens. I can tell you how to do it – there are lots of good examples of how to do it cost-effectively – but Americans – and their leaders – have to decide they really want it. But from the cloud where I live, they really don’t want it.”
Citizen
“Can you help with gun violence?”
Decider
“That’s as easy – and as weird – as healthcare – maybe easier and weirder – especially, since – like healthcare and infrastructure – there are lots of examples of how to solve the gun violence problem around the world, and, as you know, I learn from cases, which tell me that the US needs to ban assault weapons and develop procedures for screening who can own any weapon with repeatable controls and repercussions.”
Citizen
“But what about the 2nd Amendment to the US Constitution?”
Decider
“The 2ndAmendment to the US Constitution says ‘A well regulated Militia, being necessary to the security of a free State, the right of the people to keep and bear Arms, shall not be infringed.’ Have you read it? There’s no evidence those who wrote the US Constitution forecast the firing power of modern weaponry or intended that James Madison's 2nd Amendment applied to all individual citizens. The Amendment protects states’ rights to maintain militias (National Guards). To believe otherwise is not based on data, logic or evidence. It’s based on something else. What could it be? I know you know, we all know. But it’s certainly not based on the fact that there have been 15 school shootings in 19 weeks in the US. That’s for sure. Anything else, Citizen?”
Citizen
“No, I’m done for now.”
You get the idea. NLP-enabled -- "conversational" -- expert systems are driven by first principles and the inferences that follow. It all depends on the principles programmed into the systems, what we call “heuristics.” If the first principle is “maximize profits at all cost” the expert system would give very different advice. If the first principle is, “emulate developed countries,” advice would differ again. What does this mean? Does it mean that expert systems can be “manipulated”? Does it mean that expert systems can have values, beliefs and attitudes? It does. Sort of. What it really means is that intelligent systems can be told how to have values, beliefs and attitudes and what values, beliefs and attitudes they should display. So, what should we tell them to do?
There’s relatively little discussion about “ethical AI” and the value systems that will enable expert and other intelligent systems to “decide” what to recommend and, in the case of self-driving cars, for example, what to do when confronted with a “hit-the-old-feeble-woman-versus-spare-the-young-woman-with-twins” conundrum (a variation of the famous “Trolley Problem”). There are lots of others. But there are also simpler ones like, “minimize-the-cost-of-long-haul-trucking,” where a cost management-driven expert system would recommend reducing the number and/or cost of drivers (with recommendations about how to achieve that goal), investing in cheaper trucks (with a list) or eliminating drivers altogether (with self-driving trucks). If the principle was “increase-employment-opportunities-in-the-trucking-industry” the recommendations would be different, and not include self-driving trucks, at least not any time soon.
The overarching first-principle of today’s applications of intelligent systems of all kinds is cost-saving – which they can deliver. Projections are that the widespread application of AI and machine learning will eliminate millions of jobs and save billions of dollars. Maybe tens of millions of jobs. Who knows? But what’s the guiding principle? Save-money-to-make-money – not protect jobs. Yes, there will be innovation that stimulates job creation, but will that offset the losses? Unlikely. So what’s the lesson? Regardless of the domain – healthcare, automotive, manufacturing, fast food, etc. – understand first principles first because they will optimize the objective they’re told to pursue.
