How AI will »talk« to us


Bandwidth of »talking«

Have you ever wished that you could project the image you have in mind (or what you »imagine«) into the mind of someone else? Because it is too difficult to say what you mean?

There is a good reason for that wish: the huge difference in the information bandwidth we are able to receive with our eyes 1 and the information bandwidth we use by the narrow language channel. The former is in the range of 10000000 bit/s, where you’d fully use the capacity of modern internet connections or the classic ethernet speed2. The latter, however, could be easily satisfied with early internet technology, like the analog modems. Say we can process something like 5 words per second, the average word has 5 characters, a character has 8 bits - then you end up with 200 bit/s. This means by wishing to be able to project an image, you are actually wishing for a 50000 times higher bandwidth.

Visual communication: up to 50000 times higher bandwidth.

Not a stuck pixel.. Language bandwidth as the area of a red square compared to the potential bandwidth of vision as the area of the gray square.
Language bandwidth (red) compared to visual bandwidth

Evolution is yet to lead us to biological 5G communication with Gbit/s rates. The importance of the bandwidth limited language communication channel can’t be overstated. Actually, this seems to be entangled with the emergence and characteristics of consciousness. Put positively, it is amazing how efficiently we can communicate, how little information has to be actually transmitted to convey complicated content. Either way, the fact that we need to encode what we want to communicate in a selection of acoustic signals or characters requires lossy compression. What the sender means to say is a result of integration of much information. The conscious parts of that information is only a small fraction of the relevant information. It is compressed by lossy encoding to fit the narrow language bandwidth. The recipient has to decode this information using subjective interpretation, which inevitably leads to a somewhat different understanding compared to what was sent3. This is not to say that we always misunderstand each other dramatically. But it means, that we usually only talk about things, where we can understand each other! Complex information with many dependencies must often be omitted, because we lack the means to communicate them.

Language Communication
Information bandwidth and flow of information in language communication. The sender has to use lossy compression to reduce the information bandwidth to a conscious state ((())), before it can be communicated over the narrow language channel to another person. The recipient has to decode this information using subjective interpretation, which inevitably leads to a somewhat different understanding compared to what was sent3.

Visual communication

In some cases we have found a good way around: visualization. As mentioned above, the by orders of magnitude higher information bandwidth allows us to achieve much better communication about complex topics. Not only can we transmit much more information, we have dedicated “hardware” which is specialized to process this information45. Have you ever experienced, how visually prepared data can totally change your »view« on complex topics6? How it can make difficult topics seem easy7? Unfortunately, a lot of effort is required for these man-made visualizations. It is not trivial to effectively use the potential information bandwidth of vision using "visual language". While we can improve also in everyday situations (Graphic facilitation), it is obviously not very realistic to expect every politician to bring beautifully visualized data for the presentation of her ideas or every math teacher to prepare the class with animated information. Mark me - this will change! We have programs coming to our aid. This is the main way AI will “talk” with us. Not only using language, using visualization! The beautiful thing is, that we only need to program parts of this capability ourselves. As said before, we are not very good at this, because we are not Chameleons and can't present. So our capabilities for live visualizations are biologically limited. AI, however, will learn to do so! Envision omnipresent, beautifully designed and animated presentations of just the right information you need, achieving an unprecedented information density.

Omnipresent and omnipresenting AI

AI using visual communication

AI can »present« instead of just »talk«. This potentially increases the communication bandwidth substantially. As a consequence, AI may »project its mind« much more effectively than by just using language.
AI communication

Omnipresent and omnipresenting AI will make two fundamental differences:

1) The amount of information processed by AI can be much higher than what we are used to. Imagine our everyday communication augmented with instantaneous information about facts and relations. Most of our bread-and-butter communication will become obsolete or at least change fundamentally.

2) The transmitted information bandwidth can be orders of magnitude higher, because AI will be able to communicate visually.

At the moment we are very focused on if and how fast AI will become “human”, when Siri, Alexa and google will talk to us like humans. It will, literally, look different. I shall be very surprised if AI just keeps “talking”. The impact is coming slowly but steadily from behind, working on our unconscious information processing, shaping the information world we perceive. This process has started already. Disruptive changes may happen, when omnipresenting AI starts transcending the language barrier. Fundamental changes have already passed before we will see AI as persons. And the person part will only be the tiniest fraction, as human consciousness is only a tiny fraction of information processing in our brain. AI will feel broader, softer than a human intelligence and somewhat intangible. But it'll still be there. What does this mean to us today? In my opinion, the evolution of AI will automatically and inevitably lead to this - no further surprise incoming.

The one thing where this will make a fundamental difference is complexity. Because complexity arises from relations, and relations are so much easier to show than to say. Politics, management, psychology, medicine and even ethics suffer from our difficulties to deal with complexity. Integrating and presenting AI will play an important role in our challenge with complexity.

1Nørretranders, T. The user illusion: Cutting consciousness down to size.user Illusion Cut. Conscious. down to size. 1991   
2Koch, K.; McLean, J.; Segev, R.; Freed, M. A.; Berry, M. J.; Balasubramanian, V.; Sterling, P. How Much the Eye Tells the BrainCurr. Biol. 2006  pp. 1428–1434 
3Küpfmüller, K. Informationsverarbeitung durch den MenschenNachrichtentechn. Z 1959  pp. 68–74 
4Rees, G.; Kreiman, G.; Koch, C. Neural correlates of consciousness in humansNat. Rev. Neurosci. 2002  pp. 261–270 
5Koch, C. Consciousness: Confessions of a romantic reductionistMIT press 2012   
6McCandless, D. The beauty of data visualizationTED 2010   
7 2018