To make this website work, we log user data. By using Shephard's online services, you agree to our Privacy Policy, including cookie policy.

×
Open menu Search

I/ITSEC 2023: 4GD smart targets to receive ChatGPT enhancement

28th November 2023 - 08:30 GMT | by Norbert Neumann in Orlando

RSS

ChatGPT could help 4GD's SimStriker avoid stock phrases and develop different personas for its smart targets. (Photo: 4GD)

The integration of ChatGPT to 4GD's SimStriker could lead to the introduction of dynamic conversational capabilities, enabling realistic interactions with trainee soldiers in diverse training scenarios.

Close combat training technology specialist 4GD has received a contract from the Defence and Security Accelerator (DASA), a UK Ministry of Defence fund, to integrate ChatGPT into its SimStriker robotic targets, 4GD announced on the first day of I/ITSEC 2023 in Orlando, Florida.

The integration would allow smart targets to conduct dynamic conversations with trainee soldiers, the company said. 4GD would work with NIAXO, a tech innovation company, to create what it claimed were varied and realistic military training scenarios.

ChatGPT would help SimStriker avoid stock phrases and help the company develop different personas for the targets based on customer requirements. It would also enable the operator to configure or select preconfigured scenarios.

IT2EC 2022: 4GD to deliver smart urban warfare training targets to the British Army

DSEI 2023: 4GD to upgrade SimStriker smart target with generative AI and computational vision

The contract was sealed in September at DSEI 2023 in London, but the companies have decided the time was right to announce the deal now.

Talking to Shephard at the training and simulation show in Orlando, James Crowley, business development director at 4GD, said the company had already begun works on the integration and started identifying some of the ‘more critical’ risks. He said that one of the risks was technological, but the other challenge has more to do with human interactions and how to manage conversations.

‘One of the key risks has always been latency and the ability of the target to assimilate dialogue, then push that through the large language model, before returning it as speech in a timeframe which is natural and meaningful to the user,’ Crowley explained. ‘Linked to that is how people manage and hold conversations.’

Crowley emphasised that this is an exploratory effort, with the aim to allow SimStiker to eventually conduct ‘fairly sophisticated’, human-level conversations. This does not mean, however, that the smart target would have to rely on ChatGPT in every scenario.

‘If you shoot at it, we’re not going to necessarily get it to say something unique, it could just use the same tokens, if you will,’ Crowley said.

4GD expects to be able to present the ‘most basic’ form of this solution at some point in the new year. The full development of the technology is expected to last about a year with three main phases.

‘There is the conversation part, the social media part and the linking [SimStriker] into the virtual environment,’ Crowley said, adding that a number of those different strands would work in parallel. ‘The technology will in be in a good place in the next 12 months…I’d like to think we’ll have a product that will be usable in a training environment. Bureaucratically, however, there’s some more work to do to make it usable.’

Crowley explained that this aspect would address the security concerns that stem from recording potentially confidential and sensitive information soldiers might say during training.

As a separate endeavour, 4GD has been exploring adding computational vision to the smart targets that would allow them to see, identify and engage with objects and trainees further away.

Shephard's I/ITSEC 2023 coverage is sponsored by:

CAE
Norbert Neumann

Author

Norbert Neumann


Norbert is the Aviation, Military Training & Simulation reporter at Shephard Media. Before joining Shephard in …

Read full bio

Share to

Linkedin