1. What experience and know-how does your company have in providing AI-based solutions for research?

In 2023, we initiated the integration of generative AI capabilities into our customer research platform. Building on this progress, the company was rebranded as GMO Research & AI in 2024, reflecting our commitment to advancing leadership in the field of AI and driving industry innovation in response to the rapid evolution and transformative impact of AI technology.

2. Where do you think AI-based services can positively impact research? What features and benefits does AI bring and what problems can it address?

AI-driven services streamline researchers' workflows, allowing them to concentrate on more advanced analyses. The integration of AI has enhanced the user experience for respondents, making the platform more accessible to new customers while significantly improving the speed and accuracy of the entire research process. Additionally, innovative approaches such as creating digital twins with AI are expected to further enhance convenience for researchers and reduce the burden on respondents.

3. What practical problems and issues have you encountered in the use and deployment of AI?
What has worked well and how, and what has worked less well and why? 

Several challenges were faced in the use and deployment of AI.

First, the accuracy of the original data is critical. Any bias or inaccuracy in the data can affect the results.

In addition, generative AI (LLM) has the problem of ‘hallucination’, which generates incorrect information. To prevent this, appropriate training data and feedback loops are introduced.

Furthermore, LLM outputs can have poor reproducibility, which can be a challenge in situations where consistent results are required. To address this, adjustments have been made to the model setup and usage.

We continue to make improvements to address these challenges by improving data quality and adjusting models.

4. Can you explain the role of AI in your service offer in simple, non-technical terms in a way that can be easily understood by researchers and stakeholders? What are the key functionalities?

Our Digital Twin offers the following benefits:

With AI acting as a ‘proxy respondent’, the Digital Twin service is completely different from existing research services, and is provided to researchers with the explicit understanding that the answers are AI-generated.

Benefits for Respondents
Even if there are multiple requests to answer a survey, the actual respondents do not feel overburdened as AI answers on their behalf. 
Even if there are additional questions or confirmatory surveys, the AI will respond to them, so there is no need to worry about losing time.

Benefits for Researchers
They can start the survey as soon as they need to, even late at night or on holidays. 
Researchers do not have to worry about respondents availability and can ask questions as many times as they wish, allowing for more detailed surveys. Also, urgent confirmation surveys can be conducted in a more flexible way.

Thus, the system reduces the respondents burden, while providing an efficient mechanism to meet the researchers needs.

5. What is the AI model used? Are your company’s AI solutions primarily developed internally or do they integrate an existing AI system and/or involve a third party and if so, which? 

Our services use proprietary AI models developed by our business partner ETE as the mainstay of our services. In addition, models provided by OpenAI are used for some of its functions.

6. How do the algorithms deployed deliver the desired results? Can you summarise the underlying data and the way in which it interacts with the model to train your AI service? 

The deployed algorithms are trained using customer data and are specifically designed to achieve targeted outcomes. 

By analyzing data provided by survey panelists, the model identifies and learns patterns, enabling more precise analysis and accurate predictions. Customer data serves as a critical input during the training process, allowing the algorithms to extract relevant features and trends to deliver results that align with customer requirements in a practical context. 

Furthermore, these trained AI services are continuously enhanced through a feedback loop, which iteratively improves their accuracy and reliability, ensuring the models remain effective and adaptable to evolving needs.

7. What are the processes to verify and validate the output for accuracy, and are they documented? How do you measure and assess validity? Is there a process to identify and handle cases where the system yields unreliable, skewed or biased results? Do you use any specific techniques to fine-tune the output? How do you ensure that the results generated are
‘fit for purpose’?

To check and validate the accuracy of our AI models, we have put in place a process whereby the results generated are checked against the original data and improvements are made. During this process, the output of the model is closely checked to ensure that it matches the original customer data and meets the intended objectives.

8. What are the limitations of your AI models and how do you mitigate them? 

AI models have inherent limitations that we actively address:

Data Dependency:
Their performance depends heavily on the quality of the input data. Biased or inaccurate data can result in biased or erroneous outputs. To mitigate this, strict data quality control measures are implemented, and diverse datasets are utilized to ensure balanced and accurate training.

Adaptability:
AI models may struggle to adapt to new or unforeseen situations, limiting their ability to handle unknown cases. To address this, the models are regularly updated and retrained to maintain their adaptability and relevance.

Hallucination:
Generative AI can generate non-existent or inaccurate information. To counter this, a robust validation mechanism is in place, involving human review and cross-referencing with reliable external data sources.

To overcome these limitations, we are continuously improving our AI models, prioritizing enhancements in reliability, accuracy, and adaptability to ensure they meet evolving needs effectively.

9. What considerations, if any, have you taken into account, to design your service with a duty of care to humans in mind?

In service design, we take human considerations into account from two perspectives:

Considerations for Clients (Survey Requesters):

Human reviews are regularly conducted to ensure the accuracy and appropriateness of AI-generated responses.
Quality control measures are rigorously implemented to deliver reliable survey results.

Considerations for Respondents:

The system blocks ethically problematic questions in advance.
Measures are in place to protect respondents' privacy and rights.
The purpose of data use (including AI learning) is clearly disclosed, and prior consent is obtained from respondents before utilizing their response data.

By incorporating these measures, we emphasize meeting social and ethical standards from the design stage and strive for responsible AI operation.

10. Transparency: How do you ensure that it is clear when AI technologies are being used in any part of the service?

GMO Research & AI clearly states in detail on our service websites which parts of our services use AI technology. In this way, we strive to ensure that users know exactly where AI technology is used.

11. Do you have ethical principles explicitly defined for your AI-driven solution, and how in practice does that help to determine the AI’s behaviour? How do you ensure that humandefined ethical principles are the governing force behind AI-driven solutions?

We have established usage regulations to prevent customers from using the service in ways that could cause ethical issues. Additionally, the AI system is designed to avoid responding to questions that are deemed problematic.

12. Responsible Innovation: How does your AI solution integrate human oversight to ensure ethical compliance?

Monitoring During Development and Operations
・QA checks are always conducted before release to ensure the appropriateness of output.
・A system is in place to continuously collect feedback from actual users in the field and apply it to improvements.

Management of Data and Usage
・The purpose of data usage is clearly communicated to respondents whose data will be used for training.
・Clients are required to strictly adhere to the terms of use.
・Outputs generated by AI are clearly labeled, and proper usage is contractually defined for clients.

13. Data quality: How do you assess if the training data used for AI models is accurate, complete, and relevant to the research objectives in the interests of reliable results and as required by some data privacy laws?

We accurately obtain information without issues by creating a digital twin for each individual based on attribute data collected from respondents, after clearly presenting the information to be obtained and securing their consent in advance.

14. Data lineage: Do you document the origin and processing of training or input data, and are these sources made available?

The input data consists of respondent's answer data, which is a proprietary information. Therefore, GMO Research & AI does not make this data available.

15. Please provide the link to your privacy notice (sometimes referred to as a privacy policy). If your company uses different privacy notices for different products or services, please provide an example relevant to the products or services covered in your response to this question.

Our Privacy notice can be found at https://gmo-research.ai/en/privacy.

16. What steps do you take to comply with data protection laws and implement measures to protect the privacy of research participants? Have you evaluated any risks to the individual as required by privacy legislation and ensured you have obtained consent for data processing where necessary or have another legal basis?

Measures to Comply with Data Protection Laws and Protect the Privacy of Survey Participants

Our AI services do not handle or process information related to personal privacy. Therefore, no individual risk assessments or consent for data processing are required.

Across our entire service portfolio, not limited to AI, we have allocated dedicated internal and external resources to fulfill our responsibilities for personal data protection and compliance with legal requirements. To address any unauthorized access or data breaches, we have implemented an incident response program supported by a designated team. Additionally, we have appointed a Data Protection Officer to ensure adherence to applicable data protection laws and regulations.

17. What steps do you follow to ensure AI systems are resilient to adversarial attacks, noise and other potential disruptions? Which information security frameworks and standards do you use?

As a procedure to ensure AI systems are resistant to adversarial attacks, noise, and other potential disruptions, all AI providers implement the necessary measures. Additionally, prior to release, the systems undergo checks by external security vendors to confirm their safety and reliability.

18. Data ownership: Do you clearly define and communicate the ownership of data, including intellectual property rights and usage permissions?

Regarding data ownership, our company ensures that the data owner is clearly identified and consent is always obtained at the time of data collection. This process allows us to clearly define and appropriately manage intellectual property rights and usage permissions.

19. Data sovereignty: Do you restrict what can be done with the data?

Regarding data sovereignty, our company clearly communicates the purpose of data usage to respondents and utilizes only data for which consent has been obtained. Therefore, the use of data is restricted strictly to the scope of the stated purpose.

20. Ownership: Are you clear about who owns the output?

The ownership of the output is clearly defined in the contract for each service.