Category: Machine Learning, AI & Statistics

A Proposal for Credit Assignment — Concerning the U.K. Goverment’s consultation on Copyright and Artificial Intelligence

The training of A.I. and machine learning models is, in the technical vocabulary of the field, a problem of credit assignment [1]. The entire achievement of a machine learning training algorithm lies in accurately tracing the myriad miniscule connections — assigning credit — to those features of the input data responsible for achieving each output target in its training data and targets.

It would be ironic then if, tasked with assigning credit to their inputs for their outputs, A.I. modellers should baulk and call that too daunting a challenge.

It is not necessary to achieve a great degree of accuracy, or timeliness or even consistency in assigning credit to content creators and IP holders whose property has been used to train models. A little effort to provide something that is half accurate most of the time would be a helpful start.

It is not necessary to invent a new compensation model, or a new process for registering the interests and contact details of IP owners. The music industry designed and implemented solutions for this problem over a century ago. Without the aid of A.I. Or a computer.

There is such a thing as legalized robbery. For instance, loan sharking used to be legal, but it was still robbery even when it was legal. We prefer to make such things illegal because it is wrong to let those with the will to do so to take advantage of people over whom they have an advantage of, for instance, physical strength.

In commerce, strength is largely financial, partly political. The bigger, better-connected company is immensely more powerful than the individual content creator. Just as with SLAPP cases and with libel laws, the legal battle is too one-sided to even contemplate.

The UK consultation on IP and A.I. provides a unique opportunity to develop and implement a far better vision. The relationship between A.I. developers and IP holders can and should be a positive, mutually beneficial, symbiotic relationship.

A.I. certainly depends, utterly, on IP creators. If the A.I. developer monetizes whilst the IP holder is left with nothing that would be parasitic and destructive[2], not symbiotic. If an algorithm and system is implemented to reasonably share with IP holders fair recompense for their works' contribution to A.I. output, the relationship could be symbiotic, mutually beneficial, and even a virtuous circle of increasing benefit to both. Not to mention the rest of us.

[1] https://www.google.com/search?q=back+propagation+as+an+algorithm+for+credit+assignment

[2] https://www.inet.ox.ac.uk/news/new-study-reveals-impact-of-chatgpt-on-public-knowledge-sharing

The Argument For Machine Consciousness

The modern argument for machine consciousness can be paraphrased as follows:

  1. We know lots about how electro-mechanical systems work.

Therefore:

  1. Let us believe that consciousness is also an electro-mechanical system (because if it is, then we will be able to understand it).

Therefore:

  1. Machines can be conscious

One can sympathise with giving belief (2) a research budget. But to confidently assert that conceivable consequences of (2) – such as (3) – are facts about the world, is bluster. Writing bluster into our legal and moral thinking would be a mistake.

I suppose the motivation for (2) is, we always want to believe we can understand things. And, impatiently, we also want to believe things can be understood well enough just with the current state of our knowledge, insights and tools.

But gung-ho optimism based on unevidenced confidence does not have a good record and can do more harm than good.

At the current time, the negative I hope we can avoid is:

Large corporations will use stories about machine consciousness and machine legal status to transfer legal risk away from themselves, making it easier for them to prioritise profitable machinery over human lives.

A example will be when a corporation denies legal liability for a robot or self-driving car which injures a passer-by, by saying it was the robot's fault, not theirs.

A better, more logical, move in the early 21st century would be to first acknowledge that

  1. The current state of our knowledge, tools and insights does not suffice to understand what consciousness is or how it works.

There is nothing that it is like to be a computation

Whatever it is like to be a bat, there is nothing that it is like to be a computation.

A computation is the abstract form. Whether embodied in pencil on paper, or in neurons in an organism, the computation is the abstract — disembodied — structure, not the embodiment.

Error-freeness per kilowatt-hour : a Proposed Metric for Machine Learning

Abstract

Accuracy on well-known problems is widely used as a measure of the state of the art in machine learning. Accuracy is a good metric for algorithms in a world where energy has negligible cost. We do not live in such a world.

I propose an alternative metric, error-freeness per kilowatt-hour, which improves on accuracy by trading-off accuracy against energy efficiency in a useful way. It has the desirable properties of approximate linearity in the relevant ranges (1 error per 1000 is 10 times better than 1 error per 100) and of weighting energy use in a way that accounts for cost of training as a realistic fraction of a delivered service. Error-freeness per kWh is calculated as e = 1/(1 + g - Accuracy)/(h + (training time in hours * (GPU+CPU Wattage)/1000)). The granularity g is the point of diminishing returns for improving accuracy. The overhead h is the energy cost of delivering a software service with no ML training. The parameters may be tuned to circumstance, but as a general-purpose metric, I propose g=1/100,000 and h=100kWh are good human scale, commercially relevant parameter values.

Detail

Where training is very expensive, it is not helpful to score machine learning algorithms on accuracy alone, with no account taken of the resources consumed to train to the level reported. In a competitive setting, it biases to the richest player; in the global, or society-wide, or customer-focussed setting, it ignores a real cost. This leads at best to sub-optimal choices and at worst to a growing harm.

I suggest that a useful, general purpose, metric has the following
characteristics:

  1. For gross errors, it is linear in the error rate. Halving the error doubles
    the score.
  2. For very small errors, improving the error rate even to perfection adds only incremental value. Perfection is only notionally better than an error rate so small that one error during the application's lifetime is unlikely.
  3. For extremely large energy consumption, the financial cost of the delivered service becomes proportional to the energy cost. We are concerned that the total human cost of energy use is very much dis-proportionate, in that emissions from increased energy consumption is an existential threat to the human race. For much less extreme energy consumption we might nonetheless accept the linear financial cost of energy as a proxy for the real cost.
  4. For small energy consumption, the energy cost of training becomes an insignificant fraction of the whole cost of delivering a software service. The parameter h represents the energy cost of a service that uses no training.

Error-freeness per kWh can now be formulated as:

e = 
    1 / (1 + g - Accuracy)
      / (h + (training time in hours * GPU+CPU training wattage)/1000)

Setting the Parameters g and h

The granularity g

For general purpose human scale and commercial purposes I suggest a granularity of g=1/100,000 is a level at which halving the error rate grants only incremental extra value. It is about the level at which human perception of error takes real effort. Consider a 1m x 10m jigsaw of 100x1,000 pieces in which 1 piece is missing. An observer standing back to see the entire 1m x 10m work will not see the error. They would have to spend effort searching the 10 meter length of jigsaw.

Changing g by an order of magnitude either way makes little different to scores, until accuracy approaches 99.999%. So an alternative way to think about g is:

“If you can tell the difference between accuracy of 99% versus 99.9%, but cannot tell the difference between accuracy of 99.99% and 99.999%, then your granularity g is smaller than 1/1,000 but not smaller than 1/100,000.”

The software overhead energy cost, h

We estimate the energy cost of an algorithm-centric software service as follows:

  • A typical single-core of cloud compute requires 135W 1 , for an energy cost of 1.0 MWh per year per server.
  • The software parts of a service in a fast moving sector (and, “being a candidate for using ML” currently all but defines fast-moving sectors) have a typical lifespan of about 1 year. (The whole service may last longer, but as with the ship of Theseus, the parts do not).
  • A typical size of service that uses a single algorithm is 4 cores plus 4 more for development and test. (Larger services will use more algorithms. We want the cost of a service of a size that uses only one algorithm).
  • 8 such cores running 24/7 for 1 year is 8MWh.

We should set h to some fraction of 8MWh. There is little gain in attempting a more accurate baseline for general-purpose use. See the supplementary discussion below. We set that fraction based on 2 considerations.

Those 8 cores are often shared by other services, both in cloud-compute and self-hosting deployments. The large majority of the world's systems—anything outside the global top 10,000 websites—have minimal overnight traffic; office-hours is more realistic. Anything from 1% to 99% of a CPU-year might be a realistic percentage, the lower figure for virtual cloud-computing and the highest for dedicated hardware.

It is pragmatic to measure training train as the time for a single training run, rather than imagining developers keep careful record of every full or partial training run during development. We can more properly account for the total energy cost of all training time by dividing 8MWh by a typical number of training runs. If the final net takes 100 hours to train, it may have taken 10 or 10,000 training runs to settle on that net, in development, hyper-parameter tuning, multiple runs for statistical analysis, comparison with alternatives and so on. For a researcher, 1000 training runs may be too little, where for a commercial team doing only hyper-parameter training and testing, 50 runs might be more than enough. It is the widespread commercial usage that concerns our metric.

Combining the shared CPU usage with a typical number of training runs, one might argue for any fraction of 8MWh as typical, from 1/20th to 1/1000th. I propose a broad-brush rule of thumb setting h= 1/80th of 8MWh, or 100kWh. For large projects, it is simple to set g and h to values based on an actual business case and costs.

Proposed general purpose parameter values

This gives us standard parameters for error-freeness per kWh of

g=1/100,000
h=100kWh

Examples

A net is trained for 100 hours on a grid of ten 135W servers, each with a 400W GPU (i.e. 0.535 kW per server), and achieves an accuracy of 99%:

  • e = 1/(1+ g - 0.99)/(h + 100*10*.535) = 0.16.
  • It reaches accuracy=99.5% by quadrupling the number of servers: e =0.09.
  • A different algorithm for the same task achieves 99.4% on the original 10 servers: e=0.26.

On a different task, a net required only 10 hours on just a single server and GPU to reach 99% accuracy:

  • e = 1/(1+ g - 0.99)/(h + 10 * .535) = 0.95.
  • It reaches accuracy=99.5% by quadrupling training time to 40hours. e=1.64.
  • A different algorithm achieves 99.4% in the original 10 hours. e=1.58.

In the first case, training costs ½ a megawatt-hour per run (around £4,000 for 40 training runs at UK 2020 energy prices) and energy cost is well-reflected in the score. We may consider the doubling of accuracy not worth the quadrupling of cost. Where the training cost is a small fraction (around £40 for 40 training runs) of the cost of a delivered service, even a small gain in accuracy outweighs a quadrupling of energy cost.

Conclusion

When you measure people's performance, “what you measure is what you get”. People who are striving for excellence will measure their success by the measure you use. By promoting a metric that takes explicit account of energy usage, we create a culture of caring about energy usage.

The question this metric aims to answer is, “Given algorithms and training times that can achieve differing accuracy levels for different energy usage, which ought we to choose?” The point is to focus our attention on this question, in preference to letting us linger on the increasingly counter-productive question “what accuracy score can I reach if I ignore resource costs.”

Because the parameters are calibrated for real world general purposes, this metric represents a useful insight into the value vs energy cost of deploying one algorithm versus another.

Supplementary Discussion [Work In Progress] – the energy cost of a software service

To ask for the energy cost of a deployed software service is like asking for the length of a piece of string. In the absence of a survey of systems using ML, the calculation given is anecdotal on 3 points: How big a service does a typical single ML algorithm serve; what is the lifespan of such a service; for what fraction of that lifespan is the service consuming power?

In 1968, typical software application lifespan was estimated at 6-7 years 2, but a single service is a fraction of such an application, and the churn of software services has increased with the ease of the development and replacement. I propose 1 year or less is a realistic lifespan for an algorithmic service in a competitive commercial environment.

The figure of 4 cores for a service arises from considering that although the deployed algorithm may only use a single core (or one low-power GPU), a service is typically deployed as part of an application with a user interface and some persistence mechanism. A whole service might then use 2 cores (for a monolithic deployment with redundancy) or 6 or more (for a multi-tier service with redundancy). Anything beyond that is likely already looking at parts of a larger application, unconnected to the work of machine learning. We can reasonably set the boundary for “that part of the system which we are only shipping because we have an algorithm to power it” at no bigger than that.

The figure of 135W for a single socket server might, in the context of efficiency-driven cloud computing, be discounted even 99% or more for low-usage services sharing hardware and consuming zero energy when not in use. Setting h=1/80 rather than, say h=1/500, probably represents very heavy usage.

Other links

On data centre power usage: https://davidmytton.blog/how-much-energy-do-data-centers-use/

1. https://eta.lbl.gov/publications/united-states-data-center-energy#page-9

2. https://mitosystems.com/software-evolution/