Why do you ask? In previous articles I explored the question of whether artificial intelligence (AI) would ever fully replace human beings. As I thought about this question it slowly dawned on me that this was not one question but a whole family of questions. Depending on your purpose and perspective what may sound like the same question becomes a complex set of different questions. It is important to know the purpose and perspective attached to the question if a useful answer is to be provided.
Some of these questions can only have one answer for the simple reason that the person asking will refuse to accept any answer but the one they knew they would insist on from the start. Other questions are unanswerable with what we currently know and perhaps with what we can ever know. However, some of the questions in this family can be answered. In some cases, we already know the answer. The challenge when an answer can be known is how do we deal with the answer?
Some will ask the question with a defined outcome. I have known people who wanted to live a certain lifestyle, sometimes a rather hedonistic lifestyle, but ethical or moral considerations challenged their choice. Their objective seemed to be to find a way to free themselves of pesky morality and justify the lifestyle they already knew they wanted to pursue.
In other cases, I have known people who have a heritage they want to protect. It sometimes seems they feel they need to make the world safe for God! They feel threatened by change. They want to protect a past way of living that they remember with nostalgia as being good and noble. For them, the only acceptable answer is one that defends the past they seek to protect.
In both these cases, the question of whether a machine can ever fully replace a human can only have one, predefined answer. My question is, why bother asking the question? You already know the only answer you will accept! Why not go to the real issue? Whether you want to justify a lifestyle or protect a lifestyle, the issues are whether your choices will prove wise in the long run and how your choices impact others. Why not talk about the real issue instead of getting into a metaphysical discussion of AI?
There is value in exploring the metaphysics related to this issue. That exploration requires us to refine our understanding of what it means to be human. It leads us to define what it means to be alive or not alive and then what it means to be human or not human. Humans have many characteristics, but which of them make us unique? There is value in advancing our understanding in these and many other topics related to the metaphysics of this question.
There are other motivations for exploring the topic. Among questions we might ask:
- We might want to know if AI will affect me? The answer is probably that it already has and will affect you more in the future.
- We might want to know if AI will become better at doing some things that I currently do well? The answer here is known. Almost certainly, as AI is developed, it will become better than humans at a variety of tasks.
- Many are interested in knowing if AI will eliminate some jobs, most importantly, will it eliminate my job? Machines have replaced humans in many jobs. As AI is developed it will almost certainly eliminate more jobs currently done by people.
- Will AI make my life better? AI certainly has the potential to improve peoples’ lives, at least some peoples’ lives. Further, AI might improve your life in some ways but have a negative impact in other ways. This leads to a lot of other questions, like how will AI affect me specifically and what can I do about it?
The list of possible questions goes on. For the questions we can already answer, those answers raise more questions, such as:
- How will AI affect my current job? The answer seems to be that any part of your current job that is repetitive or can be defined by an algorithm is a target for AI.
The jobs and tasks AI is most likely to be successful at are the mundane, repetitive tasks and more complex tasks that can be reduced to an algorithm, even if it is a complex algorithm. Let’s think about self-driving cars and the intelligent highway system. Those are extraordinarily complex tasks. It will take very sophisticated sensors and algorithms with a lot of computational power to successfully do them. It appears that current technology may well be up to the challenge.
- What jobs will AI eliminate? Any job that can be reduced to an algorithm is a target for AI.
These questions refract and multiply based on your situation. If you are a developer of AI then you are looking for subjects to apply your skills to. The future looks pretty rosy for those who can help develop AI solutions. The challenge becomes how to direct your career so that you maximize the potential benefits your skills are opening up for you. However, if you earn your living doing a job that is threatened by AI, then you want to know how to plan your career to protect your livelihood. Your job may be done better by a machine in the future, but you still need to earn a living. The questions may sound similar, but they are leading to different answers.
The jobs that AI is least likely to replace are those that are primarily focused on tasks that AI cannot currently and may never be able to do. What are the characteristics of tasks that AI may never be able to do?
- Tasks requiring character – AI cannot perform tasks that require character. Take the golden rule. An AI engine cannot recognize that it is about to do something that it would not want done to itself. Humans can design AI engines to follow the golden rule, if they have thought through what that AI engine will be doing and identified the circumstances in which the golden rule might apply. This would demonstrate the need to have a human with good character and insight involved in the design and perhaps operation of AI.
- Creative and artistic tasks – Truly creative tasks are far beyond what AI can do. AI can use machine learning to find patterns and extrapolate data but that is not truly creative, it is algorithmic.
- Dealing with non-normal distributions – Most algorithms have an unstated assumption that the data they are analyzing has a normal distribution. Often the assumption is that the distribution is gaussian, but that is only one example of a normal distribution. When the data has a non‑normal distribution AI is limited because the underlying math in its algorithms break down.
An important category of non‑normal events are what have come to be called “black swans.” A “black swan” is a low probability, high impact event. The Fukushima nuclear plant disaster and 2000 Bush v Gore election are two examples. This category of events cannot be predicted from the record of past events because no examples are likely to be in the data.
- Tasks that require relationships – A distinctive characteristic of the most outstanding human beings is their ability to develop a network of relationships to achieve some large, complex task. Whether it be the CEO of a new startup company, the general of an army or a political leader solving a societal problem, the most outstanding humans know how to develop and manage relationships. While AI machines can act cooperatively their relationships must be designed into them by people. AI cannot independently map out those relationships and define their structure. Relationships must be designed into AI by humans.
The best way to protect your career is to choose a career that primarily deals with tasks that have these kinds of characteristics. If your current job seems threatened by AI, you should consider developing new skills with the characteristics mentioned. If you are preparing for your career now, expect AI to change the job you are considering, as it is now done. In the future AI will almost certainly do some tasks that people now do and will do them better than a person ever could. However, with some insightful preparation, focusing on these characteristics will result in AI doing the boring parts of many jobs, releasing people to be more creative, productive, and because of that, more valuable.
It is useful to be precise in the question being asked. It can happen that the answer to one question is inappropriately applied to a different question, resulting in a flawed conclusion. I have concluded in a previous article that AI faces an ontological barrier that prevents it from ever truly replacing humanity. However, if I apply that answer to my personal situation, I might conclude that I cannot be replaced in my current job. The truth is that I might well be replaced by a machine and find myself out of a job. It depends on what my job is. So, concluding that from a metaphysical perspective AI cannot replace humanity says nothing about whether AI might do some jobs currently done by humans. As AI is developed it is important to understand its impact on us individually and our societies and cultures. We may well make societal decisions that we do not want AI developed in ways that have certain impacts. Monitoring how AI is developed and its potential for further development will be an important activity if we want to maximize its benefit while protecting ourselves from undesired consequences.
You may also wish to read:
Will Humans Ever Be Fully Replaceable By AI? Part I (Stephen Berger)
Will Humans Ever Be Fully Replaceable By AI? Part II (Stephen Berger)