IEEE Future Directions
Related Articles
Area 5 – Context Aware Machines
The drive towards autonomous systems requires machines to become context aware. Technology (sensors and AI) is supporting increasing levels of awareness. In the coming decades we can expect machines to increase their awareness to levels that may compare to the awareness of living beings, humans included. In certain areas, thanks to better sensory and processing capabilities, their awareness might even exceed human awareness.
Overall, the consensus is on machine reaching a high level of context awareness, in some situations exceeding the one achievable by humans.
Q 5.1
When will machine context awareness match that of an average human?
All experts indicated this goal is beyond 2050.
The matching can only happen when the richness of machine sensors becomes comparable to that of human sensors. This will take a very long time. Until then, machine context awareness will not match that of the average human. However, if we better define or qualify what we mean by this, our answers will change. For example, in very constrained situations for humans, machines may excel (human vs. machine with IR/thermal sensors in a dark cave).
Q 5.2
Will machines in the future develop their own context awareness without being pre-programmed to recognize the various components of their environment (e.g., will they autonomously become aware of the difference between a cat and a dog)?
Unanimous agreement of the expert that this will be achieved. It will require an acceleration in AI that we did not witness in the last decades but that we are starting to experience in these last few years.
Q 5.3
When a machine will be “surprised” by an unexpected context will it autonomously re-evaluate its model of the world (as opposed to today, when they are basically halting operations and transferring control to a human)?
Unanimous agreement of the expert that this will be achieved.
It will depend, obviously, on the kind of surprise the machine experiences. If it is only an alert in the functioning of the system, or an unexpected object or context in the field, there will be need for minor adjustments. In the case of a major disruption, it is probably safer in most, but immediate emergency situations (such as, for example, the oft repeated argument of a choice to be made by an autonomous car between several equally undesirable options), to let the system inform the relevant actants of the situation and shut down or put itself on hold.
This is already be explored via saliency detection and autonomous exploration algorithms that are sensitive to novelty or anomaly detection. What machines do upon being surprised is a matter of risk posture to which they are designed (halting operations, logging the surprise for later human assessment, focusing in on the surprise to collect more data and create new models of it, etc).
Q 5.4
Will machine context awareness eventually lead to machines changing their goals?
Unanimous agreement of the expert that this will be achieved. This is clearly raising crucial issues since it may no longer be possible to trust a machine to work and operate within a predefined framework.
This is happening already — a very simple low-level example being autonomous mobile robots that
encounter blockages along planned paths through an environment and have to re-plan to new sub-goals in continued pursuit of blocked goals. More sophisticated instances of this may be mere extrapolations of such simple examples in some cases.
Q 5.5
As result of context awareness a machine might alter its behaviour. Will this change of behaviour be considered as a reason for change in the context (i.e., will machines become self-aware of their relationship with the context?
Unanimous agreement of the expert that this will be achieved. This aspect, similarly to the previous one, will be raising crucial issues although of a different sort. Here a machine may operate to change her context, including those components in the context that have full autonomy and awareness, thus leading to potential fight and to the attempt to eliminate opposition.
Q 5.6
Following on 5.5, will machine endeavour to change the context as result of their context awareness?
Unanimous agreement of the expert that this will be achieved. Same considerations as for the previous question.
Q 5.7
Following on 5.5, will machines be able to see context as the result of the interplay of several components rather than seeing the context as a static situation?
Unanimous agreement of the expert that this will be achieved.
Q 5.8
As machines become better in context awareness, are we going to hand over to them our context awareness? Will AR provide/complement our context awareness leading to a symbiotic relation in the area of context awareness?
The experts split in their view, with a part foreseeing humans to hand over context awareness to machines, implicitly trusting them to be better in assessing a situation, other negating this outcome, considering that awareness will not be delegated to a machine. Notice, however, that in limited situations, such as pilots trusting the auto-pilot, this delegation of awareness to a machine already happens. Part of the experts see the future as an extension of what is already happening today, the others put a limit to what can be delegated to a machine.
