Innovation Ciphers & the ‘adjacent possible’

Adam Timlett
May 14
10 min read

To cite, please use timestamped PDF

by Adam Timett

Introduction

The idea of an innovation cipher is a type of heuristic, with a strong theoretical and principled foundation, for identifying ‘adjacent possible’ (see reference below) products that can be made from existing products.

If I was to form an anagram from the word ‘LISTEN’, then the step before forming an anagram is to write the letters in a circle, to ‘scramble’ them by freeing them of some of their original structure.

word circle

The ‘decoding’ of the scrambled (over-compressed) message, is then different to the original but conserves some of the original

investment in the original product. An example of such a descrambling different to the original here is ‘SILENT’ .

Innovation is, in this sense very like an anagram. An anagram is a new message constructed from the original message, with the letters or words rearranged. Hence, the final ‘decoded’ message is actually a message in itself, a new product formed from the same or similar parts. Therefore, there is more than one solution to an innovation cipher, just like there can be more than one anagram made from the original message. The ‘descrambling’ of the message after the scrambling phrase can have many different solutions.

The reason for calling this type of heuristic an innovation cipher, is that it is related to cryptography and certain classical ciphers.

What I mean by this is that we can think of the existing product as a ‘message’, meaning we render the existing product as information which is structured, and so can be thought of as equivalent to the original message in a cryptographic process. However, unlike cryptography, the unscrambling has many potential solutions, not one unique solution. We can create new messages this way, rather than just scramble and recover the old message. The purpose of this is for innovation not for secure messaging.

The ‘scrambling’ of the message is then an over-compression of this original message, a description of it that is in some sense so compressed that we lose some of the sense of the original message in the new message or product. This is like the circular arrangement of letters, the intermediate step in fining the new word to make from the original word’s letters.

In an anagram, this step is often also done to help solve the anagram, but it is just as useful when constructing the anagram or to find new products from existing products.

The step of scrambling and descrambling is therefore like going ‘up’ one level, and then down to a different place from where you started:

Anagram as 'adjacent possible'

Innovation as the adjacent possible

The parts are like conserved phonemes or words in the anagram. By writing the parts down so as to reflect existing investments and sunk costs, we can find different products that are ‘adjacent possible’, (a concept due to Stuart Kauffman, see reference below) to the current products. In other words, we can find innovations that we can actually make from the original products we already have, or plan to make.

In an anagram we often conserve only the letters used and are usually not allowed to use extra letters or not use all the letters, but in an innovation cipher we represent the investment in the original product by defining ‘parts’ that we try to conserve. This is like sounds made of multiple letters or one letter in an anagram. Hence, it is more like making anagrams where each part is a word rather than a letter. We think of the changing of the ordering or description of the parts as finding new ‘affordances’ for the parts than were in the original product. But we don’t scramble the parts themselves, because they represent an investment we have already made in a particular part of the original product that we don’t want to give up. That’s why we don’t scramble everything. In addition, that continuity of part structure tends to carry information about viability of the new product. By conserving parts we tend to create new products which are more likely to be useful.

Lip Reading as a first example of an innovation cipher

So, let’s consider that we have an original product, which is a machine learning algorithm for lip reading, these are the parts described a bit like an overview of the process, with inputs listed first and then outputs.

1. Lip videos.

2. Known Words said.

3. Map from Lips to Words.

4. Some accuracy of results.

What we are developing here is the mapping of lip movement to words said, so we are constructing this map based on data to train the machine learning algorithm to automatically lip read.

We want to produce something like an ‘anagram’ of this message that implies a different potential product, a different process that might also have value, using the same parts.

But there are some missing components.

Lip animation is not in the original message for instance but could be part of a potential new product based on these parts.

However, maybe if we over-compress the message we can recover a different result, a bit like writing the parts in the circle, to ‘free up’ and make ambiguous the desired structure, but we are doing this simply by re-describing the parts in a more general way. So, we end up with:

1. Moving Lips

2. Words

3. Map between lips and words

4. Some accuracy of results

So now, having scrambled the part description a little to free up the potential meaning and make it more ambiguous, a parts stable anagram from these more ambiguous parts (like a word stable, sentence anagram) looks like this:

1. Words

2. Map between lips and words

3. Moving Lips

4. Some accuracy of results

This innovation cipher now implies a process of animating lips from words. This is a product that could be useful in the world of computer-generated images, such as the film industry. This was an actual innovation used by a researcher whose PhD was based on lip-reading, but who pivoted to produce lip animation from words as a fall back when the original target proved not to be feasible.

Lip reading to lip animation adjacent possible

Notice that the concept of reversal of the mapping isn't there and it's not actually a simple reversal because three elements change position not just two. This is actually necessary to imply lip animation versus two inputs to create a map. Reversal of the use of the map is implicit. Also, over-compression of the parts is necessary to be able to view this as some meaningful new process. We move from lip videos that imply existing data to moving lips which is ambiguous. It's an over-compression of the description of the parts, freeing up that part to be actually formed in a different way. Hence, this is a heuristic based on utilising natural language and its special properties, but it’s grounded in solid principles. We know why it works and what is happening.

In biology, such over-compression is also useful to create the instability to find different uses for similar products as an organism develops. The ability to ‘innovate’ in this sense is therefore likely essential in developmental biology also as a way of managing risk in the organism and how it evolves to be able to adapt. This is referred to in my Substack article: Systems in Nature. Why they have higher plasticity & innovative capacity, why efficiency is an organisational tuning parameter

If we ask an LLM to expand on the innovation cipher then it likely gives a different idea of the meaning of the message as a process but one that maps to the innovation of lip animation. It decompresses to a different product. In other words, it’s one solution of an innovation cipher. The new product is a scrambled version of the first, like an anagram, which is just a scrambled message that happens to have meaning on its own.

Therefore, the over-compressed model, the step where we made the description of the parts more ambiguous still conserves value, primarily, the sunk costs in existing parts.

But, as a result of the over-compression, it is more indeterminate in a mapping to a product and multiple solutions are possible, just like there can be mulitple anagrams of a sentence.

Because of the parts and over compression steps it's emphatically not just combinatorics; it relies on the properties of natural language, but it could still be generalised to other systems, like biological systems where over-compression is also a model of certain behaviours and processes. Notable examples of such over-compression is given in the related ideas of Mark Isalan (see reference) and the value of over-compression in developmental biology as described in my other Substack article, already mentioned above.

Deep space survey innovation example

Here’s another example of mapping to an actual change in the use of a product. This is more complex, because in this case we need extra information to find the new product. It’s like making an anagram, but this time we need to know that outside information can be used as part of the solution.

1. Deep space survey

2. Collect extensive data on existing deep space objects in the whole sky.

3. Repeat for accuracy over a wide range of the sky

One over-compression looks like this:

1. Do space survey

2. Collect data on the whole sky

3. Revisit same parts

So, the last part of the over-compression is not really just an over compression but more like a re-wording. In the light of the actual innovation result, or innovation cipher that was used, it is key to use this particular type of re-wording rather than over-compress in a way which simply loses too much information.

This is because the actual re-use of this data was to identify solar system objects that could be found due to their change of position against the backdrop of ‘static’ deep space objects when the same part of the sky was scanned twice on succeeding days or weeks.

So, here, to create an innovation product we need another perspective with other information about this innovation cipher. It is an ‘incomplete’ innovation cipher, in this context. The complete cipher looks like this:

1. Do space survey

2. Collect data on the whole sky

3. Revisit same parts

4. Detect near-earth/solar system objects

The rewording here in part 3 is what allows us to trace a path to the actual reuse and pivot that occurred. Notice that we added a new step, Step 4, which wasn’t in the original message. A very important class of such anagrams are therefore not finished products, but unfinished, relying on information to be added exogenously. This makes them different to mere anagrams, where typically you have to use all the existing letters or words, and you can’t add any more. We can describe such unfinished products as ‘bits’ rather than complete messages, and I also describe how Nature likely has sub-systems producing a mixture of finished products and bits, precisely to allow more innovation and uncertainty leveraging exogenous processes to be more adaptive and hedge risk about the uncertainty of what products to produce.

Fund of different perspectives

So, we need a fund of different perspectives to draw on when we create the innovation cipher for realistic use in many cases, these can be used to take interesting unfinished products, and finish them by this exogenous work. This is why innovation ciphers are not really algorithms, they are data artefacts that are constructed by heuristics, but depend also on agency and exogenous activity.

James Webb Space Telescope Calibration Example

From a risk perspective the calibration step of the James Webb Space Telescope (JWST) is worth considering. The telescope represents a highly valuable resource and any telescope time is highly sought after by researchers. It also represents high sunk costs. The set up of the telescope once in position required that its instruments be calibrated by being pointed at some faint stars for a while.

We can represent the calibration step like this:

1. Point telescope at faint stable light stars

2. Calibrate by using certain lenses and filters.

In this case we can create an innovation cipher without scrambling, by just filling in more details about which faint stable light stars.

1. Point telescope at [these] faint stable light stars

2. Calibrate by using certain lenses and filters.

In this way that cipher is that a calibration process is also doing useful science for the astronomers interested in particular faint stars, and who can do research with certain instruments that also need calibrating.

It's these adjacent possible descriptions that constitute what's being explored, and innovation can be very subtle when the space to change to an adjacent possible is already there, ‘hidden’ in the original message or product.

James Webb Space Telescope model, credit: wikipedia

Parasitic Search

In the same vein, we can take yet another astronomy example, that of the Search for Extra-Terrestrial Intelligence (SETI). In the 1980’s this was not well funded, and a method of ‘parasitic search’ was used to parasitically use whatever data that funded researchers were collecting from their use of a radio telescope, to collect the data in the radio frequency spectrum relevant to the SETI researchers and log it on their own star map of their SETI search as part of a small extra step in the process. Again, it is like the calibration step in JWST a small change to an existing process. The original process is this:

1. Point telescope at these stars for reason x directed by these astronomers.

2. Collect these radio frequencies

Here we change not where the telescope points, but just what information we value from that, in addition to the main process which values different radio frequencies. The innovation-cipher here is that the process is itself not really changed.

1. Point telescope at these stars for reason x directed by these astronomers.

2. Collect [these (different)] radio frequencies.

3. Log on SETI star map.

Notice, that because the order of steps and process is hardly changed this has consequences for the efficiency of the parasitic search, it is not an efficient search process. The same part of the sky could be searched an arbitrary number of times, because SETI is not in control of the search process. However, because of this, it can also be extended an arbitrary amount of time, because SETI is not funding the process either, and doesn’t have to show results.

Like the deep space survey example exogenous information was added, but in addition, there were not any changes to the original steps, except the precise frequencies collected. This is like the example of the JWST calibration. Hence, this is itself an important type of innovation cipher where results and the new purpose is itself remains somewhat encrypted in the existing process, to form a different kind of hidden innovation, showing that the relationship between cryptography and innovation is more than skin-deep. I also elaborate on some of the relationships to cryptography and plasticity in my book ‘On the Origin of Risk’ in the technical appendix, and also in my other articles on Substack and my website, www.turingmeta.org.