The 5 Levels of Artificial Assistance

At Attractive we are building a complete artificial UX expert, and I often like to joke that the job of any programmer is to make everyone unemployed. Except I'm not really joking. A key goal of automation is to remove the need to spend effort on a task, and you will find that great coders are on a continuous quest to make even themselves redundant. This can sound spooky, and it should. Our society is currently completely incapable of dealing with a reality where most human work becomes unnecessary, with only a few at the top of the automation pyramid being able to benefit. The good news is we are still some way from that — best current estimates place the moment for the arrival of artificial general intelligence at somewhere between 2040 and 2050, and it is worth noting that AI predictions have often been optimistic. The last jobs to vanish will be those in service, caretaking and creative industries. The very essence of those jobs relate to our human experience and as a result they may never disappear. That, however, does not mean technology should not be allowed to help in those spaces as well, to remove the mundane and upkeep parts of the job.

With AI assistance businesses popping up like dandelions in spring, it's worth looking at where exactly we are on that journey towards replicating human intelligence. To do that I have grouped artificial assistance into five main levels, starting from classical automation (the very reason technology exists at all), all the way to the full ability to work in complex and changing human environments, without any guidance.

Level 1: Traditional Computing, Pre-defined Tasks. Level 2: Guides Itself, Picks Relevant Data. Level 3: Learns from Patterns, Suggests. Level 4: Interacts in Human Environments. Level 5: Passes Applied Turing Test.

Each domain can and most likely will manifest in different types of assistants. For instance, it makes little sense to spend effort in creating an AI that has perfected the ability to drive in a city environment, and to then combine that with one which is the perfect chef. The idea of sitting in a self-driving vehicle while gorging on a wonderful meal admittedly has certain appeal, but there is enough of a challenge in accomplishing even one of those tasks amicably. It would be silly to attempt the combinatory complexity of doing both with one bot. Even humans specialise.

A sumo wrestler and ballet dancer showing how even humans function differently to specialise
Image credits: Nakatani Yoshifumi (cc 2.0) & Stano Novak (cc 2.5)

It's worth noting that Level 5, or Turing+, does not necessarily mean that the AI would pass the famous Turing Test as it is commonly specified, but would instead pass an applied one for their particular domain. Again I would like to emphasise that AI assistants will specialise, and for many it makes no sense to pass a conversation-based test. Once again taking self-driving car AIs as an example: Level 5 means the ability to completely independently deal with all situations, even odd and specific ones, where they have to drive amongst pedestrians, roadworks, and adverse conditions. To give an example: one evening as I was merrily driving home I ended up being forced onto the hard shoulder by the widest transport I had ever seen. There was a service car leading it and the driver rolled down his window to explain the situation and to tell me to move well out of the way. A Level 5 assistant should manage even that, without human intervention. Today there are no Level 5 AI assistants, in any industry, and even Level 4 assistants (that show extended autonomy and situational awareness, but cannot fully replace a human in their domain) are either very rare or non-existent.

Finally I have defined Level 5 assistants as Turing+ as it would be pointless, for most industries, to recreate a human on a one-to-one basis. The AI needs to have abilities beyond a human to be worth the investment. It needs to be able to perform its job more efficiently, at greater scale or with greater knowledge than a human.

UX Assistants


The space of artificial UX assistants is relatively new. Many people even falsely believe that UX can only be understood at all by humans. However, there is decades worth of research in user interfaces, human perception and human-computer interaction. During the last decade there has been an influx of tools, including those analysing SEO, technical aspects of interfaces, accessibility etc. Let's take a look at how the Artificial Assistance Levels apply to the UX space.

Level 1: View-oriented, dumps data. Level 2: Interacts, filters data. Level 3: Learns from customer. Level 4: Interacts with team, creates new. Level 5: Full conversation on channels, indistinguishable from human

There is a long way to go to reach a full Level 5 UX assistant — one that can interact with you to find great solutions, respond to detailed queries, and which participates in planning meetings as a full team member. For the foreseeable future there will remain a role for talented human UX designers and experts, particularly in the creative, problem-solving areas of building lovable products. At the moment there are very few artificial UX assistants that could even be classified as Level 2 assistants and none that are at Level 3. At Attractive our analyses have so far been partially human-assisted (we use humans to guide the engine and as quality assurance), but very shortly we will be releasing our first fully automated version, which will allow us to reach Level 2 assistance. We will be studying the results from that carefully and collecting feedback and improving.

We intend to hit Level 3 assistance by 2021. Meanwhile, by all means go to Attractive.ai to see what we're doing now at Level 2.