Generating Trust

In his Communications The Profession of IT column, “Can Generative AI Bots Be Trusted?” (June 2023) Peter J. Denning wrote “At the time (2019), no working examples of Creative AI (Level 4) were available to the public.” Well, maybe no working deep neural networks, but the creative power of AI in the form of evolutionary computing has been known for several decades. For example, Peter J. Bentley and David W. Corne’s 2002 Creative Evolutionary System¹ contains chapters from computer-generated music to novel fighter jet combat, whilst Koza et al.’s 2003 GP4 book stresses genetic programming as a routine invention machine.² Indeed, the annual EvoMusArt conference has been demonstrating computer creativity in the arts since its inception in 2012 (see https://link.springer.com/conference/evomusart).

References

1. Bentley, P. and Corne, D., Eds. Creative Evolutionary Systems. Morgan Kaufmann, 2002.

2. Koza, J.R. et al. Genetic Programming IV: Routine Human-Competitive Machine Intelligence. Kluwer Academic Publishers, 2003; doi:10.1007/0-387-26417-5_1

W.B. Langdon, London, U.K.

Author’s response:

Langdon seems disquieted that I did not mention genetic algorithms as examples of creative AI. Genetic algorithms are rooted in simulations of evolution processes starting in the 1950s. In all my time in computing (since 1964) we venerated these algorithms as powerful methods of optimization, not AI. So of course I did not think to mention them as examples of AI. The whole focus of my column is on whether GPT is trustworthy. Genetic algorithms have proved to be trustworthy. Whether they should be considered AI is beside the point.

Peter Denning, Monterey, CA, USA

Conferences Revisited

We are writing in response to “Revisiting ACM’s Open-Conference Principle.” (Communications, July 2023), While we agree with the need for “guiding principles for conference-site selection,” and share in the mentioned concerns about reproductive healthcare, we want to highlight some additional and critical considerations: the rights and safety of trans and non-binary members of the ACM, who potentially face harassment and imprisonment simply for going to the bathroom or for dressing as themselves (in states with bathroom or drag laws); and risk not having their medical needs met in states with trans-related healthcare laws. These are immediate harms, not just potential harms should an emergency arise. Right now, individuals emboldened by these laws are violently policing bathrooms in private spaces, whether it is legal or not. Further, individuals engaging in bathroom policing do not know who is trans, and so anyone they perceive as trans and/or non-binary is a target—including any conference attendee, trans or otherwise.

A state that we believe conferences should immediately avoid is Florida. A trans person who lands in Florida is immediately at risk—doing something that we all do several times a day in our daily lives is not only likely to result in harassment: if they need to use a toilet in a government owned building (such as an airport or museum hosting a conference reception), they potentially face a class A felony and one year of prison. Trans and non-binary attendees also face concerns similar to those of pregnant people who cannot find providers willing and able to provide the care they need in an emergency. Such emergencies, while far rarer than the need to use a bathroom and dress as oneself, are potentially life threatening (in the case of pregnancy) and also important to consider in setting conference policy. Our goal in writing this is not to downplay any specific community’s concerns. Conference locations (including hybrid and online options; and which city or country a conference is held in) regularly exclude members of our community who are juggling caregiving responsibilities; are denied visas; are disabled; are trans; or are carrying a baby. Health and safety may be at risk for multiple of these groups. Conference travel is contributing to the climate emergency. In the face of all this it may seem hard to make decisions. A good starting place is to lay out basic values for conference location selection, and ensuring the health and safety of all who wish to attend is a bare minimum.

Jennifer Mankoff, Seattle, WA, USA

Richard E. Ladner, Seattle, WA, USA

Amy Ko, Seattle, WA, USA

Jonathan Aldrich, Seattle, WA, USA

Author’s response:

I thank the writers of this letter for broadening the scope of the issues I raised in my column regarding ACM conference-site selection. There is an urgent need for ACM to formulate both guidelines and a process.

Moshe Y. Vardi, Senior Editor

Communications of the ACM

Houston, TX, USA

Dismantling Scaffolding

Vinton Cerf highlights a constructive analogy if not metaphor in his April 2023 Communications column “Scaffolding as Catalyst” when building new systems or new infrastructural primitives. However, I find that challenging to apply on a granular level. A more conducive way, for me, is to dismantle the scaffolding part into a question involving “threshold developments.” For instance, we can ask, “What conceptual threshold advance enabled us to break free from the previous structure, or even the scaffolding itself?”

Cerf invited readers to respond, and two experiences from different perspectives and eras shed light on the inquiry.

The first experience is that of Ken Thompson, who worked on porting Unix to the PDP 11. The Unix kernel was initially written in assembly language, and needed to be re-implemented into the then still evolving C language, which was a step up from its precursor, the B language, due to the addition of data types. The key question here is, “What threshold advance or features of a higher programming language would enable a skillful user to re-write assembly code?” Ken struggled with this task and failed three times due to the difficulty of keeping track of many moving parts! It was only until C had implemented structures which made the building process feasible, or for that matter, humanly more manageable!

The second experience relates to infrastructural primitives in web3, specifically the evolution of fully on-chain decentralized exchanges (DEXs). The first generation of these exchanges failed in terms of liquidity on Layer 1 blockchains due to slow throughput (and potentially high transaction fees). Instead of relying on an on-chain order book and market makers to provide liquidity, the first generation of Automated Market Makers (AMMs), used mathematical formulas to instantly determine the price of an asset based on supply and demand in the liquidity pool. This allowed for the creation of more practical on-chain DEXs with higher liquidity, with Bancor having pioneered this space and Uniswap having successfully popularized the notion of AMMs.

In light of these examples, it would be interesting to hear Vinton Cerf’s thoughts on how we can identify and prioritize “threshold developments” when designing new systems or infrastructural primitives. What criteria or framework can we use to determine which features or advances are most critical to make a new system feasible or successful?

Evangelos Georgiadis, Hong Kong Science Park, Hong Kong

Author’s response:

This falls into a category of question of the form: What can I not do now which, if I could do it, would make a major difference? Thresholds are roughly speaking exactly these enabling technologies that take you to the next level.

Vinton Cerf, McLean, VA, USA