Overview:
SG Innovate, a supporter of entrepreneurial scientists, held a community event focused on the future of deep tech in Asia. Jasmine Leng, Assistant Director of Talent Networking at SG Innovate, introduced Peter Norvig, Director of Research at Google, as the keynote speaker.

Deep Tech Innovations:
Peter Norvig, a renowned expert in artificial intelligence, was invited to share his insights on upcoming deep tech innovations.

SG Innovate’s Talent Programs:
SG Innovate offers various initiatives to cultivate deep tech talent within Singapore.

Summation Program:
Apprenticeship projects in fields like AI, cybersecurity, and IoT are provided by deep tech companies. Suitable for undergraduates and PhD students seeking summer internships. Application deadline: January 15th.

Infinity Series Program:
An extension of Summation, offering projects in software development, data analytics, and UI UX. Geared toward undergraduate and PhD students seeking summer internships. Application deadline: January 15th.

PowerX Program:
Traineeships in robotics companies for individuals with one to two years of coding experience. The first batch is currently underway.

SG Innovate’s Portfolio Companies:
The Talent Marketplace platform showcases SG Innovate’s portfolio companies, providing information about their offerings and job opportunities.

Machine Learning Applications in Robotics:
The field of robotics is experiencing a surge in interest and development, similar to the early days of personal computers in the 1970s. Practical applications of robotics are emerging, particularly in areas such as autonomous cars and warehouse robotics. Automated cars face challenges in interacting with the real world, but specific applications like delivery vehicles are becoming feasible. Sensors are becoming more affordable, driving the growth of robotics and computer vision.

Computer Vision:
Computer vision plays a crucial role in robotics, enabling tasks like object tagging and image understanding. The field is exploring applications across various domains, including medical imaging and autonomous driving. Adversarial images pose challenges to the reliability of computer vision models, prompting research on understandable and reliable models.

Medical Imaging and Protein Folding:
Computer vision models can achieve performance comparable to or exceeding that of trained doctors in medical imaging. These models can detect indicators of diseases beyond the scope of human doctors’ abilities. Advancements in protein folding have opened up new avenues for designing drugs and therapies.

Large Multimodal Models:
Peter Norvig emphasizes the significance of large multimodal models, combining data, computing power, and diverse modalities. These models have the potential to revolutionize various fields, including natural language processing, computer vision, and robotics.

Recent Advancements in AI:
Larger and more complex models with numerous parameters have led to improved performance. Multimodality combines data types like images, videos, and language, enhancing the capabilities of AI models. Language models can handle various tasks like conversations, translations, and text generation, all within a single model.

Automating Machine Learning:
Automated machine learning enables users without expertise to build machine learning models with minimal effort. Systems can generate machine learning models based on provided data and desired outcomes, democratizing access to AI technology.

Challenges and Opportunities for Deep Tech Commercialization:
2020 was a difficult year for commercializing deep tech due to the pandemic-induced uncertainty and disruptions. The pandemic accelerated digital transformation and automation across industries, creating opportunities for deep tech companies. Automation has become a strategic business objective for ensuring business continuity and resilience. Commercializing deep tech requires addressing challenges such as customer hesitancy, remote deployment difficulties, and negotiating through virtual channels.

Harvesting Ideas and Promoting Commercialization:
Governments and organizations should provide funding and support for deep tech research and development. Encouraging collaboration between academia, industry, and startups can accelerate innovation and commercialization. Promoting awareness and education about deep tech can help businesses understand and adopt these technologies. Creating a supportive ecosystem with incubators, accelerators, and mentorship programs can foster deep tech startups’ growth.

Lowering the Barrier to Entry:
Peter Norvig emphasizes the importance of making it easy for customers to try out products, especially when they are expensive. By offering a low-commitment option, such as a revenue-share approach or a small initial investment, companies can attract more potential customers. This strategy shortens the sales cycle and allows businesses to connect with a broader range of individuals within a company.

Old-School Sales Strategy:
Suchitra Narayan highlights the fundamentals of sales strategy, which involve delivering what the customer wants and allowing them to grow. Once customers see growth, they are more likely to pay and share a portion of their success with the company. This approach is considered an old-school sales strategy but remains the foundation for effective sales.

Deep Tech Developments in Asia:
Sinuh Arroyo discusses the rise of software computer science companies in Asia, particularly in Singapore. Companies like BeSense and Tiger are leading the way in visual search and automation. Research centers and labs in Asia are actively involved in European research programs, fostering collaboration and innovation.

Opportunities for Cooperation:
Arroyo emphasizes the numerous opportunities for cooperation between Asian and European companies in the field of deep tech. Research programs and initiatives facilitate collaboration and the exchange of knowledge and expertise. These partnerships can lead to groundbreaking developments and advancements in deep tech.

Asia’s Focus on Smart Manufacturing and Biotech:
Asia, particularly Singapore, is making significant investments in smart manufacturing, biotech, and food innovation. There are numerous promising companies in these fields, such as Bioformis, which has received funding from SoftBank. These companies have the potential to revolutionize their respective industries and make a positive impact on the world.

Global Startup Ecosystems:
While Silicon Valley remains a prominent tech hub, other regions such as Asia, Africa, and South America have thriving startup ecosystems. Local startups have a deeper understanding of their markets and can tailor products to meet specific needs. With the right drive, ideas, and execution, startups from anywhere can achieve success.

Robotics and its Applications:
Robotics has vast potential in various fields, including healthcare, delivery, and manufacturing. The pandemic has accelerated the demand for contactless solutions, making robots more attractive for various tasks. From self-driving cars to flying drones, there is a wide range of possibilities for robotic applications.

Robotics and Uncertainty:
Robotics presents unique challenges due to the unpredictable and uncertain nature of the real world. Robots must learn to adapt and navigate uncertainty, which opens up possibilities for solving complex problems. Reinforcement learning and exploration techniques are essential for robots to learn and operate in the real world.

Sim-to-Real Conversion in Robotics:
Robots can learn faster in simulation environments than in the real world due to the ability to run experiments at accelerated speeds. Sim-to-real conversion allows robots to transfer knowledge gained in simulation to the real world, improving their performance.

The Simulation-Reality Connection:
Simulations improve as more data is acquired, leading to exponential speed-up in learning. Robotics acceleration is expected once simulations reach a certain level of accuracy.

Pandemic-Driven Shifts in Robotics Adoption:
Safer distancing requirements slow down supply chain logistics, making robots more feasible. The pandemic has accelerated the acceptance of automation as a necessity.

Overcoming Inertia in Digital Solutions Adoption:
Reducing labor costs and increasing operation reliability are key drivers for automation. Freeing human intellect from repetitive tasks allows employees to focus on value-creating aspects of their jobs. Automation should offer employees better opportunities for training, self-development, and career progression.

AGI and Data Training:
Peter Norvig emphasizes the need for effective generalization in AI, moving away from supervised learning that heavily relies on high volume data training. The goal is to enable AI models to learn from existing knowledge and apply it to new problems with minimal additional data. Multimodal models are promising in this regard, allowing pre-trained models to be adapted to specific tasks with smaller datasets.

Robotics and Funding Mechanisms:
Peter Norvig highlights the longer timescales involved in robotics prototyping and time to market, similar to physical product development. Startups in robotics should adopt strategies used by hardware companies, such as building prototypes, finding customers for components, and generating revenue streams while working towards the final solution. Robotics startups should not feel singled out in facing these challenges, as they are common in many industries that create physical products.

Human Survival and Key Skills in the AI Era:
Sinuh Arroyo emphasizes the importance of continuous learning and adaptability in a dynamic environment where jobs are constantly evolving. Individuals should keep an open mind, embrace new skills, and progress in their careers by continuously acquiring knowledge and skills. Peter Norvig stresses the value of being a person and understanding human needs, as robots are not as capable in this area. There will always be jobs for individuals who understand human factors, even in an AI-dominated world. Commercialization of technology often favors solutions that integrate well with human users, rather than those that are technologically superior.

Technology Perfection and Customer Satisfaction:
Peter Norvig emphasizes that aiming for perfection in technology is not always necessary. Providing 80% of the desired solution can be sufficient to address a customer’s needs effectively.

Use Cases for Companion Robots:
Companion robots for kids and seniors can be successful even without advanced technology. Human willingness to meet technology halfway and engage with it plays a significant role in their acceptance.

Google’s Plans for Asian Deep Tech Startups:
Google offers programs to support Asian deep tech startups, including training, consulting, and access to various technologies.

Personal Tech Predictions and Wishes for Deep Tech in 2021:
Norvig highlights the importance of defining the intended purpose and goals of technology. He believes that optimization techniques are readily available, but the challenge lies in determining what we truly want our technologies to achieve.

Technology Focus Areas for 2021:
Robotics: Advances in robotics are expected to continue, with a focus on improving dexterity, autonomy, and collaboration with humans. Computer Vision: Ongoing progress in computer vision is anticipated, enabling more accurate and versatile object recognition and scene understanding. Natural Language Understanding: A significant breakthrough in natural language understanding is predicted, allowing for more natural and comprehensive conversations with devices.

Challenges in Building Trustworthy AI Systems:
Fairness and Bias: Experts emphasize the need to address fairness and bias in AI systems to ensure they are beneficial to all users and avoid perpetuating discrimination. Privacy and Transparency: Concerns about privacy and transparency in AI systems are highlighted, calling for measures to protect user data and ensure accountability.

Quantum Computing and AI Chips:
Quantum Computing: While still in its early stages, quantum computing is seen as a potential game-changer in problem-solving and could have a significant impact in the coming years. AI Chips: The term “AI chip” is questioned, as experts suggest that specialized hardware for AI tasks may not be necessary for all applications.

Emerging Fields with Potential:
Smart Manufacturing and Biotech: The pandemic has brought attention to the importance of smart manufacturing and biotech, leading to increased investment and research in these areas. Food Innovation: The need for food independence and sustainability drives interest in innovative approaches to food production and agriculture.

Quantum Technologies:
Quantum Sensing and Signaling: In the near term, quantum technologies are expected to make an impact in areas such as quantum sensing and signaling, rather than full-fledged quantum computing.

Challenges in the AI Chip Industry:
Lack of Standardization: The AI chip industry faces challenges related to a lack of standardization and the need for specialized hardware for different AI applications.

Terminology: AI Chips:
Peter Norvig suggests that the term “AI chip” is too limiting, as it fails to capture the versatility of these specialized chips.

Specialized Chips and Parallel Computing:
The exploration of parallel computations with specialized chips offers significant advantages. Power consumption and battery usage become crucial factors, opening up possibilities for edge computing.

The Revolution in AI:
The recent advancements in AI have been driven by data, algorithms, and the increase in computing power.

Data as an Asset and Liability:
Data’s value as an asset is acknowledged, but concerns about privacy and stewardship responsibilities have emerged. The concept of federated learning empowers users to retain their data and run processes on their devices.

Measuring Intelligence:
Peter Norvig proposes a two-by-two breakdown to define intelligence: Thought processes (validity and coherence) vs. results (achieving the correct answer) Human-like operation vs. achieving the best possible result

The Turing Test and Defining Intelligence:
Norvig interprets Alan Turing’s Turing test as a call to define intelligence through task success rather than human imitation. The emphasis is on the task itself, not the task of imitating a human. Norvig believes Turing would have embraced new tasks as measures of intelligence.

AI Limitations:
Software’s ability to play games like chess or AlphaGo is not a true representation of intelligence. AI works well when applied to specific, limited domains.

Intelligence Definition:
Intelligence should encompass a wide range of human abilities, not just isolated tasks. Software can be considered intelligent when it demonstrates the versatility and creativity of a human mind.

Predictions for Tech Trends:
Increased adoption of automation and robotics solutions, becoming mainstream. Digital transformation is no longer optional in the post-COVID remote work world. Virtual collaboration systems need improvement, especially for casual conversations and idea exchange.

Call to Action:
Connect with the panelists on LinkedIn or reach out to SG Innovate for further discussions.