If you are into AI and/or Technology it seems that every week there is a new and groundbreaking Neural Network architecture or AI paper, some even come with amazing demos or new startups using said tech and while all this is wonderful we still don’t have a sentient/conscious AI.

👋 Hi there 👋 all my content is free for Medium subscribers, if you are already a subscriber I just wanted to say thank you ! 🎉 If not and you are considering subscribing, you can use my membership referral link, you will be supporting this and other high quality content, Thank you ! ⭐️ Subscribe to Medium ! ⭐️

As a researcher in the topic I believe in attacking the unknown from various angles and so this is a mostly lighthearted look at the requirements for sentience applied to the humble toaster.

Sentience vs Consciousness : Both terms are used interchangeably but there is disagreement and variety on the actual definition of either, to some sentience is the capacity to experience feelings and sensations whileconsciousness is the awareness of internal and external existence but you could mix and match these terms and still have a valid (or invalid !) definition for either, these are also fairly shallow definitions but we can use them and add to them as we go along...

A Sensational toaster

The first thing we would need is for our toaster to sense the environment as this is the bedrock of sentience ( we start sensing in the womb) and opens our first can of worms, what should our toaster be able to sense ? We humans are very visual beings (about half of our cortex is devoted to visual tasks) but just to be on the safe side let’s also give our toaster other senses like hearing, temperature ( probably important for a toaster ) and smell, why not.

If only my toaster had a brain

We would also need a place to plug in all this sensors, so some form of electronic brain would be handy.

We can leave the hardware side (for now) and focus on the software side.

Is sentience a software or hardware problem ? Our brains make do with biological hardware that is organized in functional systems which we haven't been able to both figure out completely and recreate artificially in hardware, so the best next thing is to recreate these systems artificially in software, it is unknown if sentience can be solely software based (I say it's highly probable) and if we can recreate biological hardware (also highly probable with time), but we are really at the infancy stage, so it is probably better to take a let's see what we can do with what we have approach...

Artificial Toaster Neural Networks

If all you wanted was for your toaster to experience sensations we’d mostly be done but feelings and awareness of internal and external existence will require something more: dedicated systems which we can start organizing via neural networks, here comes another tricky part as our biological neural networks serve a purpose (survive and reproduce) so we presumably need to also give our toaster purpose, let’s split the bill and give it the following one: survive and toast bread.

Now the most basic neural networks in your brain usually deal with perception and action, which if we are to believe evolution serve the purpose of survival and reproduction (for the most part); still missing from our toaster are complex actions, it can only toast bread that has been carefully placed in its slots, but that’s all the action we need for now to make our first toaster related neural network that acts on this perception…

But how does the toaster know about toast ? Well, it's complex but for now let's just say that in humans and other species knowledge is a mix of innate or instinctual programs and acquired or experiential bits of information, which means that for now we need to program the perception and recognition of a piece of toast and the corresponding action.

Remember me ! Toaster.

We could add more and more action/perception pairs to our toaster and by and large that is how we make things “smart”, but if our toaster is ever going to be sentient it will need advanced forms of memory, specifically the ability to make new memories or learn on it’s own, a lot of hand holding is probably still needed but a decent place to start is face recognition ( babies do it ), so let’s add some Machine Learning to our toaster neural net…

Reality Check: Up till this point everything we've discussed is achievable with current off the shelf technology (software/hardware) and easyish to find know-how, there's even some AI-Smart-Toasters out there... but alas we've reached a cliff, what follows is unexplored territory and while these are my best guesses based on a few decades of research they are still bound to be incorrect or lacking, there's also a considerable effort needed to test these theories in both hardware and software, caveat emptor!

— — — — — — — — — — — — — — — ✁ — — — — — — — — — — — — — —

Powers of association & the executive toaster mind.

Much of the higher cognitive abilities we humans posses are believed to be an expansion of the perception/action associations we have discussed along with memory in all its forms (mainly working, short term and long term) these expansions then form dedicated systems, some of which might not be strictly required for sentience depending on your particular definition but some are clearly (at least to me) relevant for self awareness.

Memory wise our toaster would need to not only acquire and learn new information, but also be able to replay and associate these memories internally, if there’s a step where we give our toaster a mind of its own this is it, at any moment then the contents of our toasters brain would include the moment to moment perception of the environment along with already experienced memories.

The next big upgrade as far as systems go is the association of memories and action programs which might just be the ticket to self awareness, ( you’d still need the preceding/following elements and content though ) I say it might because a complex idea like the concept of self is believed to be created by multiple associations of other simpler concepts related to the self (things like stuff out there and stuff that is not out there that I can or can’t control), as secondary proof our cortex is almost entirely involved with associating inputs and outputs (ie long term memory, imagination, dreams).

There’s one last general piece to the sentience related systems puzzle and that is the need for executive functions in our toaster which amounts to a dedicated bridge between long term memories/programs and the current moment to moment perceptual contents, if you remember the face recognition task, executive functions happen between the perception and action stages (what to do when X is perceived), but the key difference here is that the system gets expanded and directed from within at will which introduces a new tricky concept: free will.

Free will toaster ? Rather than start with a philosophically difficult term like free will, we can go lower in the definition and just start with autonomy, that is the toaster is left to do its thing by itself, the next rung up would be something like volition which amounts to the toaster freely choosing between options at its disposal that serve its goals (survive and toast), obviously there's not much in the way of choices and survival is equally scant (beyond overheating the toaster doesn't need to worry about  much else as it can't do much about it either) and thats perhaps the trickier thing about sentience, by some accounts it needs to be experienced in a richer way, so on to the last step...  

Give your toaster a good life !

Let’s say we have all the above systems in place, we will probably need to expand them in capacity and modalities but this will still not generate sentience since we are missing content, and not just any kind of content…

A suddenly sentient Toaster ? The odds of a very advanced Artificial Neural Network or Software/Hardware system suddenly becoming sentient I believe are slim to none, I go into some detail here but the gist is that sentience/consciousness is so far proving to be a complex system that needs to be understood (we are halfway there ) and built on purpose (we've barely started), while our current AIs are for the most part very specialized and task driven, but as they increasingly become better at mimicking human behavior there's bound to be a lot of false positives.

From the moment we are born we are exposed to a rich world where there are other beings like us and a few thousand years of knowledge,customs and well civilization, some if not most of this rich environment is needed for the concept of self we usually expect from sentient beings, so it follows that in order to make your toaster sentient you will need to expose it to a rich environment, fortunately we already have one but we would need to provide equally rich environmental sensors…

The virtual alternative: When we train a neural network to recognize animals we usually don't take our computers to the zoo, we simply show them pre categorized images virtually, so there is the possibility of creating a virtual environment where AIs might become sentient, unfortunately we would also need to recreate the richness of the real world as well as the hardware interactions.

Speaking of complex content we humans (and other sentient species) are not born alone, even before our birth we are exposed and depend on others for our survival, the contents of our world if measured by an outside entity would overwhelmingly be filled by human related things such as feelings and traditions which mostly make sense in a species setting, it’s unclear if we can translate this content without the proper context or if we might need to create a toaster society to go along with a sentient toaster…

How to know if your toaster is sentient ?

In works of fiction like sci-fi movies and books the newly sentient toaster would scream “I’m alive !”, in reality sentience will probably happen very gradually and we’ve yet to give our toaster a voice (add that to the list).

The alternative to self reporting (ie asking the toaster for a full account of its sentience like humans do) are tests, there’s a variety of them, for my money though the mirror test could be a good starting point :

The Mirror Test (as applied to infants and modified for potentially sentient toasters): surreptitiously place a mark on your toaster and place it in front of mirror, if your toaster touches the mark it is a sign of self awareness (what is this thing on my toaster body)...

Alternatively as we are the ones building the sentient toaster we can also peek into the mind of the toaster so to speak, something we can’t yet do in humans and other animals.

Some parting words

If you followed all the above steps, filled some of the gaps, the unknown unknowns, and of course built it, congratulations you should have a sentient toaster by now to keep you company and do toaster related stuff !

I’d be remiss if I didn’t mention the ethics of bringing sentient life into our little corner of the universe but to be honest we are not precisely an ethics loving species are we ? What with war, inequality and such ? I still hope and believe that we will mostly be benign parents to artificial sentience if we manage to create it, there will be misuses for sure, but I want to believe the net sum will be a positive for humankind.

I should also mention that while we covered the big things in sentience we barely scratched the surface as much of the implementation needs to be done from scratch and things like emotions, general knowledge and general intelligence ( which are also part of sentience/consciousness) are deep topics, but I hope this gave you a decent idea of the problem and challenges.

Thanks for reading and toast on !