Open Source vs. Proprietary AI: A Tussle for the Future of Artificial Intelligence
The approach to AI development is only getting mired in the same tussle that influenced the development of computing in the 20th century.
- Artificial intelligence (AI) is plausibly the most exciting technological breakthrough of the 21st century so far.
- However, the approach to AI development is only getting mired in the same tussle that influenced the development of computing in the 20th century.
- Is open-source AI development the right way, or is the proprietary philosophy the way to go?
Last week, IBM and Meta joined hands to launch the AI Alliance, further broadening the gap between the philosophy of artificial intelligence or AI development. The two companies, alongside over 50 other founding members, are advocating for an open-source future for the rapidly growing technology.
IBM and Meta’s collaborators and partners include AMD, Intel, NASA, CERN, Hugging Face, Oracle, Linux Foundation, Red Hat, Harvard University, and other educational and R&D institutions.
AI Alliance Members
By definition, software development projects open to the public for use, modification, or distribution, allowing engineers, developers, etc., to collaborate is known as open-source. Open-source “fosters community collaboration and transparency, accelerates innovation, and reduces development costs,” Ravi Narayanan, global practice head of insights and analytics at technology consulting company Nisum, said.
Open-source AI encompasses training corpus, training corpus cleaning and preparation, code used in training, trained models, inference code, and guardrails code on output, explained Garry M. Paxinos, CTO of netTALK CONNECT and NOOZ.AI, to Spiceworks News & Insights. It also includes the platforms, tools, datasets, and application programming interfaces (APIs).
The Divide Between Open-Source and Non-Open-Source AI Dev
AI models and underlying hardware are possibly the hottest AI properties in town today. Considering open-source models are nowhere near as advanced and capable as private ones, the list of companies notably missing from the AI Alliance, OpenAI, Microsoft, NVIDIA, Google, DeepMind, Amazon, Anthorpic, Tesla, and multiple other AI bigwigs, is telling of the open-source and non-open-source divide.
OpenAI CEO Sam Altman and former OpenAI chief scientist Ilya Sutskever were asked by a person in the audience during a discussion at the Tel Aviv University in June this year whether open-source large language models (LLMs) can match GPT-4 without additional technical advances.
“Am I wasting my time installing Stable Vicuna 13 billion-plus wizard/ Am I wasting my time, tell me?” asked Open Source AI researcher Ishay Green, leaving Altman at a loss for words and Sutskever speechless for 12 seconds. Here’s what Sutskever answered:
“To the open-source vs. non-open-source models question, you don’t want to think about it in binary black and white terms where, like, there is a secret source that will never be rediscovered. What I will say, or whether GPT-4 will ever be reproduced by open-source models — perhaps one day it will be, but when it will be, there will be a much more powerful model in the companies. So, there will always be a gap between open-source models and their private models. And this gap may even be increasing this time. The amount of effort, engineering, and research it takes to produce one such neural net keeps increasing. And so even if there are open source models, they will be less and less produced by small groups of dedicated researchers and engineers, and it will be from the providence of a company, a big company.”
Well, sure. A strong financial backing can help companies attain a technological headstart and thus a competitive edge, Nate MacLeitch, founder and CEO of QuickBlox, opined. Sundeep Reddy Mallu, SVP at Gramener, assessed that there is “at least a 3X gap between open-source and closed-source AI models today. AI model building benefits from access to large computational power, immense granularity of data, and minimal guardrails on what can be done with it.”
Still, open-source models can play at their strengths, Narayanan said. “Open and closed-source AI models each have their strengths, typically excelling in distinct areas due to their inherent characteristics and approaches. The technological gap between these models varies: open-source models often lead in innovation and community-driven improvements, while proprietary models may offer unique, specialized capabilities and robust support.”
See More: AI: The Wakeup Call to Improve Open-Source Software Security
The AI Alliance
“Meta and IBM are spearheading the AI Alliance to drive standardization and ethical frameworks in AI, leveraging their expertise and resources. This aligns with their goals of shaping AI’s future, ensuring their influence in the evolving landscape, and fostering trust in AI technologies,” Narayanan said. “For Meta, it’s about integrating AI more deeply into social platforms and digital interactions, while IBM focuses on enhancing its enterprise AI solutions and services.”
Meta, a company considered a money-minting enterprise with a disregard for user privacy, is now at the forefront of open-source AI development. However, for a company pioneering open-source, it is bizarre that Meta requires developers/users to submit download requests and asks for details such as dates of birth for its Llama 2 model.
To Meta’s credit, the download link arrived in my inbox within minutes of registering. Perhaps Meta’s past is its biggest enemy to be skeptical of the company’s intentions. Further, Meta has made its licensing process so easy and quick that Llama 2 looks like an open-source model when it can’t really be called such, considering its dev is off-limits to the public. So, it is questionable why Meta is spearheading the AI Alliance.
Meta’s positive effect on the spur for open-source AI development can also be considered accidental. “One can see the usefulness of open-sourcing the trained models by looking at what happened after the Meta Llama models were leaked and then subsequently released by Meta formally as Llama 2. Once the trained models were leaked, there was an explosion of open-source projects and models that either used llama and/or fine-tuned the models,” Paxinos added.
Meta and IBM’s embrace of and contribution to open-source could be a part of their “goal to challenge the largest players in generative AI and to create an alternative ecosystem of AI-related companies and tools,” MacLeitch said.
Reddy Mallu concurred. He believes the aim of organizations in the AI Alliance is twofold:
- Create alternatives to commercial AI models. He believes open source has eventually come on top, as the history of software development indicates.
- The participating organizations want to differentiate themselves from early leaders in this space.
“I personally have a skeptical view of alliances. While they can be useful and beneficial, I have been on several technical committees where very large companies pay senior staff to participate, with the main purpose of slowing down the committee’s work. I’ve chaired some sub-committees where this has happened.” – Garry Paxinos, CTO of netTALK CONNECT and NOOZ.AI.
The Case for (and Against) Open-Sourcing AI Model Development
Private AI development and models can be detrimental to innovation. Jennifer Chayes, dean at UC Berkeley’s College of Computing, Data Science, and Society, noted, “Pursuing open innovation levels the playing field, allowing everyone to share in the benefits of generative AI.”
MacLeitch told Spiceworks that flexibility, the ability to customize and modify according to the needs, and the fact that they are peer-reviewed and, thus, offer higher security are the most significant advantages of open-source AI.
Narayanan added, “Open-source AI is a catalyst for innovation and accessibility, breaking down barriers for smaller entities and fostering a collaborative environment for rapid technological advancement. It offers significant cost advantages, reducing development and operational expenses, and promotes transparency, which is crucial for ethical AI development and building trust in AI systems.”
The benefits of generative AI, or AI in general, are a crucial aspect of organizations’ efforts to boost productivity, gain a competitive advantage, and devise innovative new products and services for the end user. However, it also accompanies deep-rooted concerns about the dangers of AI technology, including its impact on consumer privacy, its tendency to create biases and discriminate cybersecurity, and the lack of clarity on its interaction with humans.
The White House’s Executive Order on AI use notes open-source models, citing them as dual-use foundation models whose weights are publicly available. “When the weights for a dual-use foundation model are widely available — such as when they are publicly posted on the internet — there can be substantial benefits to innovation, but also substantial security risks, such as the removal of safeguards within the model,” the Executive Order reads.
Secretary of Commerce Gina Raimondo is expected to submit a report to the president by July 2024 on policy and regulatory recommendations after consulting with the private sector, academia, civil society, and others on what constitutes potential benefits, risks, and implications of open models.
“There is a heightened potential for misuse, including ethical concerns and societal harm. Open-source AI projects often grapple with inconsistent quality and maintenance challenges, impacting their reliability. Additionally, they pose significant security vulnerabilities and complex compliance issues, particularly in intellectual property and licensing,” Narayanan continued.
Specifically, MacLeitch explained that “Open-source AI algorithms can be used to create deepfakes and other tools for online scams, in addition to the spread of disinformation. At the extremes, open-source AI could be used to create autonomous weapons.”
Paxinos goes on to point out the reason why the dangers of AI are inherent to the technology.
“There is a deeper philosophical issue with these dangers. Many of the dangers are really psychological. We are concerned that models may present any number of biases in their outputs. And while those biases are of real concern, in many respects, they reflect our history. Are we losing the ability to understand biases for what they are and thus learn from our mistakes? At the same time, depending on the domain, understanding these biases may help us make better decisions – especially when working in an adversarial environment.
While it is a worthy goal to be altruistic, we also have to be realistic about human nature and deal with it appropriately. And also ensure our ‘guardrails’ don’t create hidden conflicts within our AI systems.”
The snail-paced development of guardrails or legal provisions on AI development and the liabilities associated with it in the U.S. and other parts of the world has fed uncertainty to this emerging field. AI developers and companies have called for AI regulation and have offered to participate in the process.
That raises another question — can their participation influence the process and tilt the regulation in their favor?
See More: The State of AI in Cybersecurity 2023: A Comprehensive Analysis
Influence on AI legislation
One way or the other, AI legislation is bound to happen. Organizations are ensuring they can steer the boat toward their interests.
“The AI Alliance can be expected to play a significant role in shaping AI legislation. As multi-billion dollar corporations working alongside high-profile universities, the Alliance certainly had the financial resources and political clout to influence policy,” MacLeitch noted.
Narayanan added, “The AI Alliance, with its collective expertise and industry clout, can significantly influence AI legislation. By providing informed insights and recommendations, they can shape policy frameworks, ensuring regulations are both technologically informed and aligned with industry capabilities and needs. Their involvement can lead to more balanced, effective, and innovation-friendly AI regulations.”
On the other hand, Paxinos expects AI regulation through legislation to stifle innovation. Moreover, he questions its broad-level applicability, whether organizations engaged in open-source or proprietary AI development.
“The question is which ‘actors’ will follow the legislation and which will not. Will it make countries that follow the guidelines fall behind the developments of countries that do not follow them?
When dealing with guardrails, who decides what is ‘safe’ content and what is not? Is it as arbitrary and capricious as how misinformation and disinformation are qualified? How is the concept of Free Speech affected? Looking at the press and publications near the time of the country’s founding, it is clear that misinformation was fought with better information, not censorship. When does ‘opinion’ cross over to misinformation? Are Thought Crimes possible?
At a deeper level, at what point does an AI have the Right to Free Speech and Free Expression? Interesting times….” – Garry Paxinos, CTO of netTALK CONNECT and NOOZ.AI.
Proprietary AI development – benefits
Despite the opaque nature of AI development that has been the norm so far, proprietary AI development does offer some benefits, including:
- Protecting intellectual property
- Can provide a better user experience
- Easy investment opportunity
- Controlled development
- Consistent quality
- Organizational alignment with goals
Do you prefer an open-source or proprietary approach to AI development? Share with us on LinkedIn, X, or Facebook. We’d love to hear from you!
Image source: Shutterstock
MORE ON ARTIFICIAL INTELLIGENCE
- Generative AI: Exploring Capabilities and Potential of Seven ChatGPT Alternatives
- The State of AI in Cybersecurity 2023: Insights About Use Cases
- Generative AI Translation: Proceed with Caution
- Microsoft Ignite 2023: Is Microsoft the Most Clear-Headed Company on AI?
- Unpacking the Straining Relationship of AI Companies and Websites Over Data Scraping
