Rethinking Technology Readiness for AI: A Practitioner's Perspective

As a technology transfer officer at Mohammed VI Polytechnic University (UM6P), I've had the privilege of witnessing firsthand the rapid evolution of AI technologies. This experience has led me to recognize a critical gap in our industry: the need for a more tailored approach to assessing the readiness of AI technologies.

Technology Readiness Level (TRL)?scale was first employed by NASA in 1974 to evaluate the maturity of technologies for spacecraft design as part of risk assessment. It was demonstrated that transition of emerging technologies at lesser degrees of maturity results in higher overall risk. Later, the TRL scale was adopted by the US Department of Defense (DOD), the US Department of Energy (DOE), then by the European Space Agency (ESA). Nowadays, The TRL scale is widely used across industries and academia.

TRL is a measure used to assess the maturity of a particular technology. It ranges from TRL 1 (basic principles observed) to TRL 9 (actual system proven in operational environment). For software, TRLs are used to evaluate the progression from initial concept to full deployment in a real-world environment.

TRL approach proved to be useful as a tool for:

1. General understanding of technology status;
2. Risk assessment and management;
3. Decision making with respect to technology funding;
4. Decision making with respect to technology transfer.

The Accelerating Pace of AI Adoption

When we look at the historical data for time-to-market in information technologies, we see a clear trend of acceleration. The internet took about 30 years from concept to widespread adoption, while smartphones achieved the same in just a decade. Today, AI technologies are poised to shatter these timelines, with potential for even faster integration across industries.

Here's a rough timeline for some prominent information technologies from concept to widespread adoption:

- Internet: ~30 years (ARPANET in 1969 to widespread commercial use in late 1990s)
- Wi-Fi: ~15 years (First IEEE 802.11 standard in 1997 to widespread adoption in early 2010s)
- Smartphones: ~10 years (First iPhone in 2007 to widespread global adoption by 2017)
- Cloud Computing: ~10-15 years (Amazon Web Services launched in 2006, widespread enterprise adoption by late 2010s)
- 4G LTE: ~8 years (First commercial deployment in 2009 to widespread adoption by 2017)        

It's important to note that the time to market has generally decreased for more recent technologies due to improved infrastructure and faster global communication.

AI's Imminent Impact Across Sectors

The ripple effects of AI are already being felt across diverse sectors. In healthcare, AI is revolutionizing image analysis and disease diagnosis. Financial institutions are leveraging AI for market predictions and fraud detection. Manufacturing is seeing AI-powered quality control and supply chain optimization. From retail to transportation, education to legal services, AI is not just coming—it's here, and it's transforming job roles at an unprecedented pace.

Here's a mapping of industries and jobs that are likely to see increased AI adoption:

a) Healthcare: Radiologists (image analysis), Pathologists (disease diagnosis), Drug researchers (drug discovery and development)

b) Finance: Financial analysts (market prediction, risk assessment), Fraud detection specialists, Customer service representatives (AI chatbots)

c) Manufacturing: Quality control inspectors (computer vision for defect detection), Supply chain managers (predictive analytics), Robotics engineers (AI-powered automation)

d) Retail: Inventory managers (demand forecasting), Marketing specialists (personalized recommendations), Customer service representatives (AI chatbots)

e) Transportation: Logistics planners (route optimization), Autonomous vehicle developers, Traffic management officials (smart city applications)

f) Education: Tutors (personalized learning systems), Curriculum developers (adaptive learning platforms), Educational content creators (AI-assisted content generation)

g) Legal: Paralegals (document analysis and research), Contract reviewers, Legal researchers (case law analysis)

The AI-TRA Approach: What's Different?

Traditional Technology Readiness Assessment (TRA) frameworks, while valuable, often fall short when applied to the nuanced world of AI. That's why I've taken the initiative to adapt these frameworks, creating what I call the AI-TRA approach. Today, I'm excited to share this adapted framework with my LinkedIn network in the hope that it might prove useful to fellow practitioners in the field.

This adapted framework retains the familiar 1-9 scale of Technology Readiness Levels (TRLs) but redefines them to better reflect the unique development trajectory of AI systems. For instance, in this framework, TRL 3 involves implementing prototype algorithms in standard frameworks like PyTorch or TensorFlow and conducting small-scale experiments on synthetic or limited real data.

I've also introduced AI-specific Critical Technology Elements (CTEs) as a cornerstone of assessment, including:

1. Core AI/ML Algorithms
2. Data Processing Pipelines
3. Model Architectures
4. Training Infrastructure
5. Inference Deployment
6. Integration Components

Putting It into Practice: The Tesla Case Study

To test this adapted framework, I applied it to one of the most talked-about AI technologies today: Tesla's autonomous driving system. The results were fascinating and highlighted the complexity of assessing such advanced AI systems.

Using the AI-TRA framework, I found that Tesla's autonomous driving technology currently sits at an overall TRL of 6. However, this single number doesn't tell the whole story. Individual components of the system showed a wide range of maturity levels:

• The Perception System (responsible for understanding the vehicle's environment) reached TRL 7, demonstrating strong performance in various conditions.
• The Decision Making and Path Planning component, crucial for determining vehicle actions, is at TRL 6, reflecting the ongoing challenges in handling all possible traffic scenarios.
• The Vehicle Control System achieved TRL 8, showing its capability to execute complex maneuvers based on AI system outputs.
• The Sensor Suite and Fusion, along with High-Definition Mapping and Localization, both reached TRL 7, indicating advanced capabilities but with room for improvement in extreme conditions.

This assessment underscores the complexity of autonomous driving technology and the challenges in achieving full autonomy (TRL 9). It also highlights why Tesla, despite being at the forefront of this technology, still requires driver supervision for its Full Self-Driving (FSD) beta.

For those interested in the detailed breakdown, I've posted the full TRA for Tesla's self-driving cars below.

Sharing is Caring

I'm a firm believer in the power of open collaboration. That's why I've decided to share this adapted AI-TRA framework with the community. You can find the templates, including guides for CTE Identification, TRL Assessment Criteria, Risk Assessment Strategies, and Documentation Requirements, at the list bellow.

• Comprehensive AI Technology Readiness Assessment Guide
• Comprehensive TRL Assessment Calculator for AI Products/Services

This comprehensive guide provides a structured approach for conducting thorough and technically precise Technology Readiness Assessments for AI systems. It reflects current industry practices and addresses the unique challenges of evaluating AI technologies across various stages of development and deployment.

These are by no means perfect, and I'm sure many of you will have valuable insights to add. I encourage you to use, adapt, and improve upon these templates. Your feedback and contributions can help refine this framework, making it more robust and useful for the entire AI community.

Why This Matters

In my role at UM6P, I've seen how crucial accurate technology assessment is for decision-making, whether in research direction, investment, or industry collaboration. This adapted framework offers more than just a numerical assessment – it provides a structured approach to identifying risks, from data drift and scalability challenges to regulatory hurdles and ethical considerations.

Looking Ahead

As AI continues to advance, our methods for assessing its readiness must evolve too. This adapted AI-TRA framework is just a small step in that direction. I'm excited to see how it might be further developed and applied across different sectors.

I'd love to hear your thoughts and experiences. Have you faced similar challenges in assessing AI technologies? How do you approach technology readiness in your work?

Let's continue this important conversation and work together to bridge the gap between cutting-edge AI research and real-world application.

#ArtificialIntelligence #AI #TechnologyTransfer #TTO #Innovation #TRL #OpenSource #AutonomousDriving

Notes:

All images were created using Leonardo.ai , with the exception of two copyrighted figures from "TRL: from aerospace to mainstream" and the Technology Readiness Levels animation.

Publication date: August 14th, 2024