Linux is arguably software developers’ favorite OS. Over 14,000 contributors have invested countless hours in developing the Linux Kernel. With Linux becoming increasingly popular due to its security and flexibility, developers who are interested in artificial intelligence (AI) may want to explore the possibilities within the Linux environment.
AI can easily be tagged as the future of technology, even if we already see it at work today. Virtual assistants such as Siri for Apple, Cortana for Microsoft, and Alexa for Amazon, are just some of the real-world examples of AI at work. The healthcare industry also uses AI in health monitoring, prescription management, drug discovery, and clinical documentation. Marketing benefits from AI as well, particularly in discovering trends, boosting revenue, and demand forecasting.
As AI becomes more and more ingrained in our daily lives through consumer products, we can’t help but be concerned that proprietary software will comprise the market. And we are not talking about a million-dollar market, but a bigger one that may reach US$118.6 billion by 2025.
Many industries and end-users would thus benefit from more open-source AI projects and tools for developers’ use. That would save tons of individuals and companies money to build their own AI-powered apps.
In this post, we explore five open-source AI projects or tools that are compatible with Linux and delve into the pros and cons of open-source AI and AI in general.
Five Open-Source AI Projects for Linux Users
The Google Brain team created TensorFlow. Its underlying software powers some of the technologies that Google uses today. It translates languages, improves search engine results, recognizes pictures in Google Photos, and understands spoken words, making its machine learning (ML) capabilities genuinely awe-inspiring.
To the surprise of the tech community, Google open-sourced TensorFlow, making it available to everyone. Developers can create ML models, classes for these models, and write imperative forward passes with it, among others. TensorFlow uses Python, C++, and CUDA.
Microsoft Cognitive Toolkit
Researchers at Microsoft Artificial Intelligence and Research initially developed the Microsoft Cognitive Toolkit, formerly known as “CNTK,” as an internal tool to speed up their research. It later served as an exhaustive toolkit for deep learning. It was first used by Liebherr to develop smart refrigerators and other appliances and powered Microsoft’s flagship products.
Since becoming open source in 2016, the toolkit has been used by different organizations to perform a wide range of deep learning and ML activities. Microsoft Cognitive Toolkit uses C++ and Python. What sets it apart is its scalability. It can train and examine deep learning algorithms in a central processing unit (CPU), graphics processing unit (GPU), and other environments.
Acumos AI is a product of the collaboration between TechMahindra and AT&T. It is an open-source AI platform that allows developers to build and deploy AI-powered applications. The platform also enables them to share AI-powered apps, fostering a community that does not hoard knowledge.
The most significant contribution of Acumos AI to the market is that it allows for easy framework integration. Integration does not need to be performed by advanced programmers since the AI platform makes it uncomplicated for anyone. It supports several software languages such as Python, Java, and R.
IBM Almaden Research Center developed SystemML in 2010 to simplify the process of scaling ML algorithms written for small to big data. Before its development, data scientists who wrote ML algorithms using R or Python would rely on system programmers to convert the algorithms for big data using a different language.
SystemML automatically scales an ML algorithm using a Python- or R-like language, effectively getting rid of the multi-iterative process, which took weeks to complete. It wasn’t until June 2015 though that IBM open-sourced SystemML, and in 2017, it became an Apache Top-Level Project.
OpenNN is a neural network library written in C++. Data mining algorithms are present within its library, which can be embedded in other software to enable developers to perform predictive analysis. It’s important to emphasize that OpenNN is inherently a software library, and so doesn’t have a user interface (UI). The library, however, powers some predictive analytics tools such as Neural Designer, which allows users to model data through neural networks without needing to code programs.
OpenNN’s development started in 2003 and was initially funded by the European Union (EU) under the research project named “Risk Assessment and Management of FLOODS (RAMFLOOD).” Artelnics, a tech company based in Spain, is currently developing the project.
Open-Source AI: The Pros and Cons
One factor that drives applications’ creators to release their work for free is the desire for AI to progress at a faster pace. By making their apps open-source, they can pool the knowledge of millions of experts together, and development becomes a lot faster as a result of this global collaboration. Between the five open-source AI applications and libraries detailed above, developers can program AI-powered software that could potentially change the world one industry at a time.
AI in Agriculture
Food security is a global issue, and with the increasing population, new methods of food production are much needed, and AI technology has been very helpful in this regard. Several countries around the world are benefiting from smart farming technologies that aid in livestock and crop monitoring, irrigation, weather forecasting, and overall farm management.
AI in Marketing
Big Data is a huge part of AI, and one of the industries that needs massive amounts of information is the marketing industry. AI has helped marketing professionals anticipate consumer demand, discover new trends, and personalize products and services. All these capabilities help companies improve their bottom lines.
AI in Healthcare
The field of medicine is also increasingly making use of AI technology. For instance, AI systems are used to monitor a patient's intake of prescribed medication. AI-powered health monitoring apps are also helping patients and doctors keep track of their heart rate and other vital statistics.
AI in Cybersecurity
Machine learning (ML), which is a component of AI, is changing the world of cybersecurity in terms of threat investigation and incident response. AI-powered cybersecurity tools can detect indicators of compromise (IoCs) such as malicious emails, URLs, IP addresses, and even unnatural network traffic.
Open-source AI is also being explored in developing hardware, specifically microprocessors that are more secure. While advances in software have become a trend, hardware is lagging behind somewhat, making it easier for cybercriminals to gain access to microprocessors. But, with the help of AI, better and more secure chips can be developed.
AI as a Cybercrime Weapon
Along with the positive undeniable contributions of AI comes the other side of the coin, though—a new generation of cyberthreats backed by AI and smart technology. Blackhat hackers are, for instance, developing malware that uses AI to circumvent antivirus and antimalware detection tools. Cybercriminals use AI to hide malicious code in benign applications by training the malware to wait until the preset triggering action is performed.
IBM Research demonstrated how DeepLocker can be used in cyber attacks. In the demonstration, DeepLocker can be trained to:
- Create an email that bypasses security filters
- Create a target profile
- Mutate to bypass antivirus or antimalware programs
- Perform cyber attacks at machine-like speed
In the future, we may even see AI-powered malware trained to recognize a target’s face or voice. As a result, current cybersecurity tools and infrastructure may become obsolete.
Aside from the possibility of cybercriminals launching AI-powered attacks, the very core of an AI machine is actually vulnerable to attacks. Threat actors, for example, can cause the deep neural networks of a system to cause it to make mistakes with the addition of subtle inputs. This vulnerability led IBM to develop an AI security software called Adversarial Robustness Toolbox (ART), which it also released as open-source software.
AI is all the rage in different industries, and rightly so. AI-powered tools and systems have the potential to change processes for the better—healthcare becomes more factual than intuitive, increases in revenue can be seen more clearly in marketing efforts, and food security becomes a reality rather than a dream.
However, we should not discount the fact that AI can also be weaponized, empowering the wrong people. Cybersecurity systems must also be upgraded to counter AI-powered cyberattacks. And when developing AI-powered machines, it is critical to ensure that they are not vulnerable to attacks.
About the Author
Alexandre Francois is a serial entrepreneur and tech enthusiast who believes that knowledge about innovations and emerging technologies should be easily understandable and available to everyone. He is also the publishing director of Techslang — a tech awareness resource where cybersecurity and IT are explained in plain English.