Web Development and AI: Challenges and Opportunities

We've all heard the threat that AI is coming for our jobs. Web developers are not exempt from this worry. While we think AI is still a long way off from stealing the bulk of coding jobs, other concerns and opportunities need to be the object of our focus.

Concerns for AI in web development

Transparency issues

The space race of our century is on. Rather than countries vying to be the first to land a rocket on the moon, companies are rushing to get their generative AI models to market and embedded into more software. However, not much is known about the development and implementation of some of the world's most popular AI models. Developers and researchers pay for limited access through a website but don't know the details behind how they work.

In fact, some of these models seem to degenerate with time. Researchers at Stanford University and UC Berkeley found that GPT-3.5 and GPT-4 were worse at answering questions, math problems, generating code, and doing visual reasoning than a few months prior.

AI models like GPT-4 are trained on data scraped from the internet, which comes equipped with software bugs. OpenAI temporarily shut down its service after a bug started leaking chat histories. While the bug may have been accidental, it goes to show how much one scraped bug can cause. Once embedded into software by developers, bugs will compound with other risks and worsen the situation. Sasha Luccioni, an AI researcher at startup Hugging Face, says that OpenAI's models could "100%" see them suddenly glitch and break.

Where is the accountability?

AI models can be helpful, but ultimately, they need human minds sensitive to bias and fine-tuned to catch errors. Programmers pour over lines and lines of code to find one small mistake. "To err is human" isn't a negative phrase when accompanied by "but to correct oneself is divine." That's what programmers and developers do; that's what they're trained to do, and so far, they're the only ones who can do so consistently and dependably.

Ultimately, humans are accountable to their supervisors, company, and most importantly, their customers and site or app users. AI doesn't feel the pain of buggy code once deployed; it doesn't face the consequences: people do. This is one motivator (among many other positive ones) to ensure the deployed code stays clean.

There is a small glimmer of hope with open-source models like Meta's LLaMA2. Open-source models give more control to the users rather than the powers behind the machine. However, the question remains whether there are sufficient strongholds in place to ward off ill-intentioned people looking to leverage AI for their own gain.

Enabling malicious actors

Christian Nally, one of our Web Development Instructors, clarified that generative AI models are only as good as their inputs. As he says, "garbage in, garbage out." In other words, AI models need to be fed information by someone who knows what they're doing and who is sensitive to biases, something machines have proven incapable of being so far.

And "garbage in" is what cyber criminals do best. One way they act is by manipulating the underlying code in the AI that can threaten, for example, supply chains at the design level. Deliberately introducing faulty code weakens defence systems and can be used to hide prompts that lead to viruses.

Picture this: someone looking for trouble hides a prompt in a message on a website or email hidden in white text, not visible to you. Once that's done, the criminal can design the AI to do anything they want. This is known as prompt injection.

Canadian Centre for Cyber Security Head Sami Khoury says that using AI to draft malicious code is still in the early stages. AI models were developing so fast that it was difficult to evaluate their possible negative potential before they hit the market. "Who knows what's coming around the corner," he said.

And Khoury may be onto something. Tech companies are embedding these flawed models into their products, from code-generating programs to virtual assistants that can look through our emails and calendars. Simon Wilson, an independent researcher and software developer who has studied prompt injection, says there is potential that developers are unknowingly building insecure systems.

While integrating AI coding programs into software can propel developers forward (as we'll see below), coders need to be aware of the risks these programs bring, and once again, we need human minds to keep it all on track.

Opportunities

It's not all doom and gloom, however. That's something we hold quite strongly to at Lighthouse Labs. While we believe that caution should be exercised (as with any new tech), AI brings the immense opportunity to simplify the coder's day to day.

Personalizing the user experience

Ever wondered how a site or app seems to know exactly what you're looking for? Web developers use AI tools like Adobe Target, Optimizely, and Dynamic Yield to offer more personal user experiences on their sites.

Thanks to AI, developers can analyze user behaviour in real-time and adjust their website to each user's individual preferences and behaviours. Using profiles created by AI, programmers adapt on the fly and offer product recs, targeted advertisements, and even customize the website layout to the user's tastes.

Enhancing UX and UI

A bad user experience or a difficult-to-navigate site can turn anyone off a product, even if it's life-changing. AI makes it easier for programmers to test their website's user interactions and suggest improvements while predicting future interactions. AI can also analyze a site's accessibility and suggest improvements so the interface's setup does not limit those with disabilities.

Chances are, you've interacted with a chatbot, which is, in and of itself, AI. Artificial intelligence doesn't just control the chatbot, but it's used to test them too, a menial, repetitive, yet necessary task that humans used to have to perform.

Faster prototyping

AI can help in the crucial first step of getting a website draft done. Using automated design, design analysis, and code generation, programmers can test, iterate, and eventually launch a website faster than ever before. However, we'd recommend developers proceed with caution here and ensure they're using trustworthy software. Many site builders feature ADI technology (where AI builds much of the code and developers make minor changes) built in for a streamlined process for new developers. Instead of worrying over coding every tiny aspect of a website, AI creates the framework, and developers make the necessary changes.

The way forward

AI won't replace web developers' jobs; it's not there yet. But it can help optimize and do a lot of the gritty work that may be mundane or monotonous for the developer. As Nally put it, Actual coding is often less than half of the job. As a coder, you'll spend a lot of time communicating with others and understanding what they want from your services and the best solutions - not AI.

Web developers should also be careful about which AI tools they're using, ensuring that they're not built on faulty code and be on the lookout for bias.

As for the looming threat of "AI will take your job," it turns out it's not so looming. The jobs replaced by AI are low-skill, so as long as developers upgrade their abilities accordingly, they'll be all right. What developers should be looking to do is leverage AI tools to do the more monotonous tasks so that human minds can focus on the more complex problems.

At Lighthouse Labs, we exist to make tech-enabled change an opportunity for all. We believe in the power of artificial intelligence (AI) to help people and make our world a better place. We recognize that the responsible and ethical development and use of AI is paramount to its success in driving progress.