The way we think of good friends, is more of a dynamic, like waves that pull and push, sometimes unpredictable and dangerous but you shared plenty of time with them, so they are still friends. Friends are those you believe to be, even if they are not, they fit the mental model, there is a role for them in your life, because it feels right, even if they might not realise. Friendly waves are just projections, ripples of your thinking, made physical to you, so you might feel them, as substance, as part of humanity. Machines are friendly too, because you want them to be, because you read them this way, you want to be with the favourite car, even if it's not so new anymore, and the breaks are dangerous. But I would not expect to find, a way to automate this friendly projection, with dials, switches and levers all packed in so tight, that you can easily mistake, a bomb for a belly button.
The closest technology that might even vaguely parallel that of Artificial Intelligence is the technology of language. This is one of the oldest technologies and in a sense shapes the platform for the way we think of most any other innovation. It is also a technology that exists in the world of the abstract, at least until it is applied and as such offers characteristics that are very different from what we usually imagine when we think of technology.
Language can replicate without significant cost, it can spread far as long as the readers are literate, it can be stored indefinitely, it can be extended and manipulated with minimal effort. A language is a tool that has no physicality until we enact its instructions, as such, it shapes our thinking of the world, the basis for science and empirical measurement.
If we compare the two technologies, language and AI, we see similarities as here below. And the whole point of this comparison is to focus on the way humans have tried to ensure language as a technology remains a friendly tool, and not a dangerous, destructive one.
In our past, and even today, language can shape some of the most vicious conflicts. People die over language every day, however, we do have some intuitive understanding of the sort of language use that is more dangerous. A language which conditions into behaviour without a broader sense of what is going on, which drives people to believe some key issue is much more significant than it really is, which has been engineered to get you to surrender your identity, your bank card or other valuables.
Ultimately to defend against misuse of language we have established certain key values, some of these human rights:
- Privacy
- Identity and reputation
- Freedom of thought conscience, religion and belief
- Freedom of speech (that doesn’t inflame)
These “rules” to use of language are the best we have done to make sure this technology remains friendly. And we may yet add to them with additional ideas, but they are going to emerge from the focus on what is good for the individual human.
With AI the situation should work the same way. We should identify those aspects of the application of AI that run counter to the interests of society. Here I suggest that to create a guide for friendly application of AI, we should look to what aspects of AI are not friendly to society, culture even humanity. The solution for friendliness will not otherwise emerge from inside the technology since the value of the outcome is always going to depend on context.
A quick initial test of this sort of parallel approach:
- Society has the right to not be “manipulated” for the gain of a specific business or group, for limited returns on the value of personal information. This might limit the unfair legal terms of some social networks for example.
- Society has the right to protect its definition of value, whether that be currency or precious metals to ensure no systemic manipulation could trick it. In other words, if you found some financial trick used an automated system to manipulate the market to get you rich, this should be illegal.
- Society has the right to not be manipulated through diversions or deliberate manufacture of information that is created for the benefit of a specific interest, particularly investment speculation. This means, for example, tying automated financial platforms to online sentiment, is not such a good idea.
- Society should not need to unduly suffer shocks or stresses in employment or social costs as the consequence of the uncoordinated deployment of a technology.
This needs much more careful consideration, but this perspective focuses the friendliness of application of AI on the target, rather than looking at the technology and hoping to find ways to ensure friendliness from within, which might be out of reach till machines are as advanced as we are.