How to start an AI project for a good cause and boost your career
October 29, 2020
From text-reading applications for blind people to cancer screening with machine vision and early detection of disease outbreaks – AI can make this world a little bit better. But how to get started with your first AI for a good project. Developers and Data Scientists working on ideas that benefit society can, in the process, cultivate their skills and develop their careers. It creates opportunities to work on exciting stuff that might not be part of your regular day to day work, and what starts as a weekend project can someday become your dream job.
How to get started?
It is essential to start with a problem in mind. You can think about a challenge you are passionate about and how AI could help detect patterns, automate repeated tasks, or scaling manual work. Then, think about your expertise, whether it’s time-series analysis, machine visions, or natural language processing (NLP), and ideas that fit with what you are good at.
My “aha moment” came a couple of years ago with an unexpected twist. The goal was to use my NLP experience to build a text generation application that will learn word patterns based on news stories and generate coherent headlines. The results were very realistic, almost indistinguishable from real headlines, that I figured – if we can generate fake headlines, we might also be able to detect them. This was the start of my journey in the AI to counter misinformation space with AdVerif.ai. A year later, the initiative was recognized by research firm CB Insights among 2019 Game Changers – startups with the potential to transform the economy and society for the better.
What will you need?
Data science and back-end development skills
Getting your first AI app to work requires two primary skills: data science and back-end development. If you are a data scientist, you are probably more used to work on the modeling part offline and then work with a data engineer or a developer to integrate it into the company product online. The two main things you will need here is server coding language and a cloud service. If Python is your preferred language, then Flask should be relatively easy to start with, and there is a variety of extensions to handle standard functionalities you might want. For a cloud service, Heroku is the right choice for beginners, and AWS, Azure, and Google Cloud are also not too tricky to master. Learning these skills, how back-end code works, and the interfaces of machine learning applications with the external world will help you become a better developer and allow you to build your first end-to-end AI prototype.
For example, think about an app that lets you scan your Twitter feed or Twitter accounts of famous people and check if AI can find any fake news there. This is the idea behind our app - the FakeRank Challenge. It was developed with Flask and deployed to Azure. The Flask code connects the back-end components, including the tweet scanner and the FakeRank AI calls, and collects the results to display via the visual front-end built with HTML, CSS, and a little JS. Read more about how it works here.
A testing partner
It is beneficial to get feedback on your AI project as early as possible to make sure you are going in the right direction. As you begin to wrap up the development phase, start reaching out to people or organizations you think are a good fit for your product, such as domain experts in the relevant field. Focus on how you can provide them value and align the expectations that this is a work in progress, and you seek their advice.
Good AI vs. Bad AI
While the latest AI advancements have unleashed great powers, some malicious applications also surface from time to time. For instance, in recent years, we have been seeing examples of Generative Adversarial Networks being used to create deep-fakes and manipulations of public figures videos. Even good intentions might occasionally end up with unwanted results. One example is a model trained on user demographics to help with mortgage lending or hiring that creates discrimination due to inherent biases.
When working on your AI project, you should keep in mind pitfalls that might cause your model to be biased. The model is only as good as the data it is learning from, and inherent biases in the data will manifest into biased predictions. This is one of the challenges of misinformation detection – finding the balance between freedom of speech while curbing misleading content and promoting trustworthy information. To this end, we partner with fact-checking organizations that comply with the IFCN principles of fairness, transparency, and non-partisanship. When using deep-learning, we also invest in developing explainable models and perform regular bias auditing.
Taking on an AI challenge can be an excellent opportunity to boost your career by leveraging your strengths to demonstrate your abilities while promoting a good cause. Developing your first end-to-end prototype can be quite fulfilling, and by working with domain experts to fine-tune your model, you will learn a lot about the task. While the road might be challenging at times, if you are passionate about the cause, you will enjoy the journey and inspire others to join in making our world a little better.
About the author: Or Levi is a Research Scientist and the founder and CEO of AdVerif.ai – a startup fighting the spread of misinformation using AI.
Ready to take charge of your dev career?
Join Europe's leading job platform for software developers!