AI Tools for Productivity - Risks and Mitigations

In the short term, even the internal utilization of AI tooling poses risks that can be highly relevant for enterprises. In this article, I cover some of the primary risks affecting large categories of AI tools, along with potential mitigations that are already possible, as well as some likely near-future approaches.

Note: This article focuses only on company-internal use cases, where productivity can be increased in various ways with current-generation, productized AI tooling (e.g., GPT-4, Stable Diffusion/Midjourney, Adobe Firefly, and other use-case-specific tools). I'm not discussing end-customer, end-to-end processes, which carry significantly higher headaches and warrant a more extensive discussion covering areas such as bias, fairness, lack of explainability, regulatory risks, reliability, robustness requirements, and ethical concerns. For internal productivity use cases, these are, for the most part, not as essential.

Privacy and data protection

The risks:

As of writing this, the most well-known services are still provided from the US, which makes them immediately problematic regarding GDPR (which is why Italy went ahead and banned ChatGPT for now). In ChatGPT's case, there have also been instances of users seeing each other's data, and there's the issue with the material used for further training data. All of these effectively block the utilization of most use cases dealing with internal and sensitive data.

The mitigations and future potential:

Microsoft has been a forerunner in AI productization for a while and is very familiar with European privacy legislation. Using their existing ecosystem, they have already published plans to integrate GPT into their extensive suite of office, productivity, and cloud offerings, which are already GDPR-compliant. They will undoubtedly ensure that companies have full control over whether their internal data gets utilized in training and that GDPR controls are fully implemented. This will naturally force their competitors to respond with similar capabilities.

Another approach is to use open-source AI tools that can be run on private networks without external connectivity, ensuring that internal and sensitive data will not be leaked. The tradeoff is a significantly less productized ecosystem, requiring higher investment and skills to utilize in most cases. The open-source ecosystem is quickly expanding and will eventually mature. Stable Diffusion is already easily locally executable on the visual side, and the recently released Open Assistant seems promising as an alternative to ChatGPT.

Copyright and IPR

The risks:

There are two sides to the IPR risk: 1) company-internal IPR leaking outside (via service bugs, used as training material that happens to be reproducible for others, etc.), and 2) inadvertent usage of others' IPR (if, e.g., code completion/generation reproduces a significant enough portion of a source work directly so that it's considered copyright violation—applicable somewhat similarly to visual, textual, and other content types).

There are several pending lawsuits alleging copyright infringement simply for producing work that is of similar style as copyrighted original works and specifically that copyright was infringed by those works having been utilized as part of the training of the generative AIs.

EFF effectively argues that this should fall within fair use, and I sincerely hope that interpretation will be widely applied. However, before we start getting precedents from courts, this will remain a risk: https://www.eff.org/deeplinks/2023/04/how-we-think-about-copyright-and-ai-art-0

The mitigation:

There is little that can be done to fully mitigate the risk, as an end consumer has no way of knowing when a generative tool produces something that happens to be close enough to an existing copyrighted work.

There are a few approaches to this:

• Wait for services where the service provider explicitly carries the copyright risk.
• One variant is using AI tools that have been exclusively trained on content explicitly licensed for training purposes, such as what Adobe did with their Firefly. This effectively sidesteps the issue.
• Realize how minuscule the risk is and simply accept it.
• Mitigate by never using generative output exactly as-is, but make manual modifications or only use generative AIs in a more assisting fashion that completes your manually provided base. This naturally reduces some of the benefit, but it can be a necessary part of the process in some use cases.

Regulation

The risk:

With the upcoming impact of AI increasingly clear to legislators, it's a certainty that AI legislation and various additional regulations will start forming in the near future. The first taste of this is the EU's draft AI Act, which, in addition to covering some essential and necessary steps, also overshoots significantly by placing unrealistic requirements on general-purpose AI providers, such as full explainability of everything, understanding all possible scenarios and preventing misuse, as well as requiring full understanding and errorless training data.

https://www.technologyreview.com/2022/05/13/1052223/guide-ai-act-europe/

The AI Act also poses a significant risk to open-source AI models and tools due to potential legal liability around general-purpose AIs, not necessarily on the application side but rather for the general-purpose model provider.

https://techcrunch.com/2022/09/06/the-eus-ai-act-could-have-a-chilling-effect-on-open-source-efforts-experts-warn/

Mitigation:

There is little we can do at the individual or most company levels, except keep our ears to the ground. The impact on internal productivity use cases will likely be significantly less and will depend more on how service providers respond (if entire categories of tools end up being effectively banned).

Competition

The number one risk is waking up to this too late while your competitors start realizing productivity benefits, allowing them to move faster and faster.

Why I'm Talking About AI All the Time: Preparing for the Productivity Boom and Its Implications

Why have I been discussing AI so frequently lately? AI, more specifically the current and upcoming generations of machine learning and large language models (LLMs), represents the most substantial leap in productivity ever. This is due to its broad applicability across numerous domains and the speed at which the benefits (and downsides) can likely be realized.

I have personally been following AI capabilities development, AI alignment, and futuristic discussions on singularity for over ten years, and it is now that we're witnessing the hype actually making its way to real world productivity. Avoiding a deep dive into societal discussions here, if we manage to evade major catastrophes, it's almost inevitable that, in the distant future, most individuals will be unable to produce something generating additional economic value beyond what AI can already achieve for cheaper and better. This is a massive subject, and while avoiding catastrophes is crucial, it's also a topic where few individuals can make meaningful difference.

In the short term, the impacts at individual, team, company, and national levels will be significant, with considerable differences between those who can effectively and rapidly adopt these technologies and those who lag behind. Disruption has been a buzzword for some time, but previous waves (internet, mobile, cloud, etc.) pale in comparison to the forthcoming AI-driven disruption.

Of course, there are factors that may slow progress: legal uncertainties and risks, the manner in which various regulatory bodies exert control, and the interplay between societal responses and these factors. However, these can only decelerate progress, not halt it, despite calls for a global pause in AI development.

At the individual level, effectively utilizing the variety of AI tools available can realistically lead to productivity increases in the hundreds of percent (depending on job and level of exposure of different tasks to current AI capabilities). It's essential to note that overall productivity doesn't scale linearly with the number of people due to communication overhead. If one person could genuinely double their productivity, their total output would likely equal that of a team of 3-4 people. Although there are key personnel risks to consider, companies cannot afford to ignore these potential gains for too long.

Individual-level progress begins to reflect at the company level, and market forces will ensure that early adopters gain a significant competitive advantage over laggards. An initial study by OpenAI estimates that 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of LLMs, while approximately 19% of workers may see at least 50% of their tasks impacted (https://arxiv.org/abs/2303.10130). This is only the current generation of GPTs, with further productization just beginning.

So, what does this mean? As an individual, you should familiarize yourself with AI tools and consider how to enhance your personal productivity. As a company, you should invest in short-term AI strategy, understand how large-scale utilization can be realistically achieved in your context, and determine which risk-benefit tradeoffs make sense.

This is not your typical overhyped technology. The results and products are already here, they're real, and they're only getting better.

Colossal Change in Software Development - AI

 Software development is undergoing a colossal change - AI. I can now honestly say that GPT-4 is at the level of a junior developer (in almost any specialization, of course). The business impact across multiple sectors is going to be huge, and there will definitely be winners and losers in how extensively and quickly they manage to adopt efficient AI assistance utilization. This impact obviously extends far beyond software development - it's simply the focus of this post.

While GPT-3.5 was already useful, and I've been using GitHub Copilot on my hobby projects for a while already, actually testing out GPT-4 more seriously a few days ago really brought home how imminent an actual paradigm shift is.

At the time of GPT-3.5, I noticed that I didn't really end up using it that much due to how many errors and bugs there were. With GPT-4, the level of quality in a smaller context is sufficient that it's a very significant productivity boost. In 5 hours of hobby game project coding, I managed to create around 800-900 lines of meaningful functional code with GPT-4 assistance.

Currently, the key to efficient utilization is having the specific goal and structure in mind, perhaps fleshing that out manually along with a few instructions and TODO comments, and then letting GPT-4 create the implementation. With this approach, the implementation was often correct on the first attempt, and while there were a few issues, they more often than not resulted from my unclear instructions or me missing some key aspect of the overall structure.

The same 5-hour period also included me learning new things, which slowed progress. I estimate that with an entirely manual approach, I might have only accomplished around a quarter of that in the same time, considering there were many aspects that I wasn't already familiar with.

With Copilot X in tech preview (i.e., GPT-4 capabilities integrated into Visual Studio) and many other similar and even better productized approaches incoming in the very near future, it's very realistic to talk about practical 5-10X productivity boosts. Even when a significant part of the overall time expenditure in overall development is not simply coding, the same approaches work for many other individual tasks as well.