I am the VP of Engineering at Apriorit, a software development company that provides engineering services globally to tech companies.
getty
Businesses are rushing to enhance their products and operations with artificial intelligence (AI), making it today’s fastest-adopted technology. Generative AI promises business owners a way to bridge talent gaps and reduce employee overhead. Other AI products bring benefits like advanced analytics, automation and a highly personalized user experience.
Yet, as exciting as the AI journey may seem, there can be lots of obstacles. Whether you decide to build an AI-powered service for your customers or optimize your company’s internal processes with the help of AI technologies, here are four crucial challenges I’d like you to think about.
1. You’ll need quality data—a lot of it.
Every smart solution starts with quality data. To deliver a competitive AI product that can make accurate predictions and provide reliable insights, you first need to gather lots of relevant, high-quality data. This means clean and uncorrupted data with no errors or duplicates.
Another important requirement is that your data must be diverse and properly labeled to improve your solution’s accuracy and minimize the risk of introducing bias. Also, avoid feeding your AI any personally identifiable data to avoid ethical and legal complications. As highlighted by Funmipe “VF” Olofinladedirect in his discussion of AI challenges, privacy by design is fundamental for building a quality AI product.
There are many ways to ensure the appropriate level of data privacy within your AI system. One of the most effective approaches so far is anonymization—encrypting or erasing anything that can potentially connect data to an individual person.
At my company, we often help our clients build and improve their datasets, ensuring quality pre-processing of data. As part of this pre-processing, we thoroughly anonymize personal data to ensure legal compliance and eliminate privacy and security concerns.
2. Start looking for AI talent or grow it yourself.
Businesses in various industries expect new AI products, especially those relying on generative AI models, to help them fill talent gaps. However, the rising popularity of AI-powered solutions itself creates a high demand for relevant specialists.
Whether you want to create a general-purpose AI service or a niche, industry-tailored solution, you’re going to need a team of AI creators with strong expertise in data science, machine learning and AI development. Depending on what team you already have in place, you might need to hire new experts, upskill your existing specialists, or do both.
As growing and advancing an in-house team takes time and money, it may be wise to delegate the development, testing, and support of your AI model to an outsourcing company. Just make sure to check that your outsourcer has relevant experience with your industry and type of AI system.
My company often works with complex projects focused on industries like healthcare and cybersecurity, and we dedicate a lot of resources to advancing the skills of our AI experts. Your company can and should do the same.
3. Ensure the secure and responsible use of your AI system.
As Gen-Zers shape new forms of human–AI interaction, companies like Samsung are restricting the use of ChatGPT to prevent leaks of sensitive data. But does efficiency really have to come at the cost of security?
When building new AI products, you need to think about their security and efficiency at the inside and outside levels.
The inside level is about what the AI model is. To protect your AI system from possible security risks, using bias-free data and reliable algorithms isn’t enough. You also need to heavily test your system and employ strong encryption and access management mechanisms.
The outside level is about what your AI model does. Adversarial users may look for destructive ways to benefit from your model. Monitoring and moderating the use of AI-powered products, especially at the early stages of their lifecycle, can help detect and prevent such attempts. Curiously, one possible solution to this problem—anomaly detection—can also be driven by AI.
When building and testing new AI-powered solutions for your clients, it's important to rely on both internal coding standards and general recommendations from tech leaders like Microsoft and OWASP. However, we as a community also need to keep working on reducing the use of AI for illegal activities and encourage the growth of ethical AI.
4. Plan for the integration and future growth of your AI solution.
To create a solution that can stay competitive in the long run, you need to account for things like a lack of future-proof hardware, high reliance on legacy systems and continually changing legal requirements for AI solutions.
For your AI product to be compatible with other solutions, you may need to adjust your technology stack, skill set and even workflow. Therefore, knowing which services and solutions you want your AI system to integrate with can help you better plan your project and prevent you from draining resources.
Planning for future growth and improvements is also vital for your solution’s performance and security. As AI models tend to degrade over time, we advise our clients to periodically fine-tune their AI systems so they can effectively handle new data and tasks.
Rushing straight into adopting AI can be risky. But when you know and account for possible risks and challenges, you get to leverage the full potential of this promising technology for your business.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
