Dealing with generative large language models (LLMs) in today’s commercial environment is enough to give anyone the blues. The wave of interest and popularity surrounding ChatGPT and similar tools released by BigTech in the last two years has been of epoch making proportions. But it has already spawned an avalanche of lawsuits, given rise to some deep concerns among those running both public sector and private sector organizations and even opened some existential questions about the future of humanity.
Why the intensity of interest? Simple really, AI offers a way through the hard graft – a way of getting from A-Z without having to deal with the ‘busy-ness’ of everything in between. Whether you are a college student, a sociologist, a business person or a government agency worker the attraction of such a tool is too real and immediate for the average human to resist. Humankind seemingly has reached the peak of achievement promised by the industrial revolution: the ultimate labor-saving device.
Why the concerns? In our 21st century world, software is king. Product (not to mention service) development is becoming virtually synonymous with software development. Software, being algorithms developed and code written by humans is the key to getting ahead of the game, whichever game you happen to be in. Software, as conceived by humans, is also the embodiment of certain legal rights, namely copyright, patents and sometimes unpatented but secret information.
The proportion of open source software has increased spectacularly, especially over the last two decades, leaving us with a complex tapestry of closed software exhibiting the hallmarks of centralized corporate control and lack of transparency (i.e. no access to source code or ancillary trade secret information) and open source/open access providing full transparency as to how the program works, oftentimes also with immunity from patent assertion by those contributing to the development or using the software.
In the past, companies with internal development operations have had to monitor their developers very closely to ensure that they are aware of the differences between open and closed source and that they know precisely what – and in what combination – they are introducing into the pipeline and ultimately releasing into the market. This has meant vigilance with regard to importing code from unverified sources, from the internet or from open source repositories like GitHub. But it has also meant paying attention to the types of development tools being used in their internal development environment.
Two types of software development tools have been particularly problematic: those that inject code into the final software product such as the GNU Bison language parser tool and those that generate code (usually through algorithmic process) in the final software product. The end result is the same, the presence of unsuspected and unwanted code that can create havoc down the line.
These are still real concerns for a development team but the advent of AI tools has thrown an additional set of imponderables into the mix. Developing software using one or more of the commercially available AI code development tools such as GitHub Copilot might seem like a dream come true but the reality is that organizations who understand software and in particular open source software are putting the brakes on. Are they prudent to do so? It will probably depend a great deal on the tool and how it is made available to the user. There are open source generative LLMs and third party deployments that are poised to compete with those offered by Microsoft, Google, Amazon and Meta and will likely be more manageable than their counterparts simply because of the level of transparency afforded to the user on the all important questions; on what data (in this case principally what code sets) was the model trained and how does the computing architecture work to generate the results, specifically the machine-learning aspects of signal processing, settings, values and of course training and evaluation code.
Users and third party service providers would do well to evaluate their use of AI tools from many different angles depending on their use case:
- Will AI assisted code development bring with it legal obligations due to the way in which the tool has been trained e.g. the use of copyright protected code sets or patented algorithms?
- Who will be the owner of any new code generated by the AI tool, assuming that any rights will exist at all in the absence of human creativity?
- What level of customization will be possible to make the tool(s) fit your specific needs?
- Will skilled developers and users be able to create intellectual property rights through the prompts they dream up to obtain a particular result or to fulfill a particular objective?
- What information concerning the tool will be or become available and at what cost to assist the user in performing risk-benefit analysis?
- To what extent – in the absence of information and transparency – will it be possible to obtain contractual protections from the supplier of the tool?
- How will your access to the tool (cloud subscription vs on site deployment) affect these and other obligations? In the latter case, what will be the benefit of open access (access to studies, design of experiment and informational documentation) without open source or vice versa?
- What will be the consequence or effect of giving the tool access through non-custom APIs to your personal or private data, including corporate databases?
No one has all the answers to these questions. Hence our conclusion that those organizations and their executives insisting on the need for guardrails before permitting their workforce to use AI tools are exercising the requisite prudence expected of them by their stakeholders. As BigTech dodges the bullets (or not as the case may be) from the litigation frenzy, the extent to which the user community will become a future target for rights owners and data subjects may also become clearer.
In the meantime, the calls for regulation to avoid a domesday scenario for the human race are reaching a crescendo. There are those who suggest that regulation will curb excesses of AI power or prevent it from falling into the wrong hands but there are also many who believe that regulation has the consequence of consolidating power in fewer and fewer hands to the detriment of public benefit.
The jury is out.
Meanwhile our electronic keyboards will be taught to play the best blues ever.
Your mOSS team
Leave a Reply